PhD Recommendation Letter Sample From a Supervisor

PhD Recommendation Letter Sample From a Supervisor

A PhD letter has to show independent research, not good work. UIUC tells PhD applicants their letters "should also address your research abilities and potential", and CMU says its committee "wants to know about your ability to do independent research". Below is a full constructed sample of about 600 words from a master's thesis supervisor, with notes on why each paragraph is there.

A list of skills, a "top 1%" with no group and a warm closing line make a master's letter, whatever the heading says. Our view: the paragraph that carries a PhD letter is the one where the student was wrong, noticed it, and fixed it without being told. That is the closest a supervisor can come to showing independent research on paper. If your draft is ready, you can get a free Quick Review of your recommendation letter.

Every program page and form quoted below was checked on October 31, 2026. None of them states a cycle year for these rules.

On this page:

• What PhD forms ask that master's forms do not
• The supervisor's five paragraphs
• Sample: master's thesis supervisor to a CS PhD
• Adapting it for physics or a UK or Australian PhD
• Co-authored work: saying who did what
• Before you submit the letter

What PhD forms ask that master's forms do not

The master's question is "can this student do the coursework?" The PhD question is "can this student run a research project?" The programs say so in their own words.

Three things repeat across the six: independence, research potential rather than performance, and a comparison with other research students. Sydney's form asks the referee to "compare this student with others known to you who have been accepted in previous years or are now applying for admission." Stanford's form asks for "the most selective group to which you are comparing the applicant" and "Approximately how many people are in this group". A master's letter can get away without a comparison group. A PhD letter cannot.

The supervisor's five paragraphs

A thesis supervisor has more to say than any other referee. The risk is saying it as a list. Use five paragraphs, each with one job:

1. Contact. Your role, the project, the dates, and how often you met. The referee section of Sydney's form opens with "In what capacity, how well, and for how long, have you known the applicant?" Answer that first.
2. The student's own contribution. What the student decided, built or found, separated from what the group or you did.
3. A setback and the response. The wrong hypothesis, the failed method, the bad data, and what the student did next. This is the independence evidence.
4. The comparison. A named group with a size and a window: "the 30 master's thesis students I have supervised since 2018". Use the same group as any form tick.
5. The endorsement. For the named program and research area, in words a committee can quote.

The sample below adds two short paragraphs a thesis supervisor often needs: who did what on co-authored work, and one candid limit.

The countercase: if the student's thesis went smoothly and there was no real setback, do not invent one. Use the paragraph for the hardest decision the student made on their own, such as choosing a method you had doubts about. Independence is the point; the failure is only one way to show it.

Sample: master's thesis supervisor to a CS PhD

Constructed example, written by us, not a real applicant statement. Context (invented): Dr. S. Menon supervised "Ananya Rao" for her M.Tech thesis at an Indian institute and is writing for her application to US computer science PhD programs. All names, places and numbers are invented. The letter body below, between the salutation and the sign-off, is 595 words.

Dear Members of the Admissions Committee,

I am writing to recommend Ms. Ananya Rao for doctoral study in computer science, with a focus on machine learning for medical imaging. I am an Associate Professor in the Department of Computer Science and Engineering at the Western Institute of Technology, where I lead a group of eight graduate students working on learning from limited labeled data. I supervised Ms. Rao's M.Tech thesis from July 2024 to June 2026. We met weekly for two years, and I also taught her in my graduate course on statistical learning, where she placed first of 41 students.

Her thesis asked whether self-supervised pretraining on unlabeled chest X-rays could reduce the number of labelled images a hospital needs to train a reliable classifier. The question was mine; the design of the answer was hers. She chose the three pretraining methods to compare, wrote the full training pipeline herself, and negotiated access to a second, independent hospital dataset so that we could test whether results held outside the hospital that supplied our main data. That second dataset was not in my plan. It became the most useful part of the thesis.

The most important thing I can tell you about Ms. Rao happened in her third semester. Her first results showed pretraining cutting the labels needed by more than half. I was ready to write it up. She was not. She noticed that the improvement was largest on images from one scanner model and checked whether the model had learned to recognize the scanner rather than the disease. It had. She designed a test that removed scanner information, reran every experiment, and found that the true gain was about a third of what we had first measured. She then presented the corrected result, and the reason for the correction, at our group meeting before I had read her notes. That is the judgement I look for in a doctoral student: she distrusted a result she wanted to be true, and she found the flaw herself.

The corrected work was accepted as a workshop paper with Ms. Rao as first author, me as last author, and one other student who contributed the data-cleaning scripts. The experimental design, the scanner analysis and the writing of the paper were hers. I edited two drafts.

Since 2018 I have supervised 30 master's thesis students. Ms. Rao is the strongest of them, and the only one who changed the direction of a project for a reason I had missed. Among the 41 students in my statistical learning course in 2024, she had the highest final mark. On the forms I am completing for her applications, I am placing her in the top 5% of the master's thesis students I have supervised over these eight years.

She has limits that a doctoral program will fix. Her mathematical background is applied rather than theoretical, and she will need to take a graduate course in optimization. She has not yet taught. As a teaching assistant in my course she held weekly problem sessions for about twenty students, and those students rated the sessions highly in our department survey, but she has not designed a course.

Ms. Rao is ready to run a research project of her own. She can frame a question, design a test that could prove her wrong, and act on the answer. I recommend her for your doctoral program in the strongest terms I use, and without reservation. I would gladly take her as a doctoral student in my own group if she chose to stay.

Please contact me if I can add anything.

Sincerely,

Dr. S. Menon Associate Professor, Department of Computer Science and Engineering Western Institute of Technology

Why each paragraph is there

• Paragraph one answers Sydney's "capacity, how well, and for how long" before anything else: role, dates, weekly meetings, and a second setting (the course).
• Paragraph two separates "the question was mine" from "the design of the answer was hers". That split answers CMU's question about the student's own "ability to do independent research".
• Paragraph three is the setback. It shows the independence that CMU, Toronto and Stanford ask about, without the word "independent".
• Paragraph four handles co-authorship: who did what, in one sentence each.
• Paragraph five names the group (30 thesis students), the window (since 2018) and the size, and matches the form tick to the letter.
• Paragraph six gives a candid limit, which Stanford's prompt invites with "Please write candidly". A limit the letter explains reads as honesty; an unexplained hedge reads as doubt.
• The close is quotable and adds a fact only a supervisor can offer: "I would gladly take her as a doctoral student in my own group".

Adapting it for physics or a UK or Australian PhD

Experimental physics. Replace the scanner story with a lab decision: a calibration the student questioned, an instrument they learned to run alone, or a measurement they repeated because it disagreed with theory. The independence question is the same.

If the letter goes to a European PhD call, check how the call gates references first: references for PhD positions in Europe.

UK research degrees. Warwick suggests asking the referee "to comment specifically on your ability in the area detailed in your proposal". That wording assumes the applicant has a proposal, so add one paragraph on the proposed topic: why the student is equipped for that topic, not just for research in general.

Australian research degrees. Sydney uses its own form rather than a free letter, with a five-point research-potential scale ("Very strong" to "Nil") and percentile bands from "up to 40%" to "90% plus". Copy the five paragraphs into the form's text boxes and make sure the ticks match your comparison sentence. Our sibling page on Australian referee reports walks through that form.

Co-authored work: saying who did what

If the student's work became a paper, the reader will ask whose it is. Be exact:

• Name the author order and every contributor's role in one sentence each.
• Say what you did. "I edited two drafts" is more credible than silence.
• Do not claim the group's results for the student, and do not hide a co-author who did real work. A later reader can check the paper.

If you are writing jointly with a postdoc who supervised the student day to day, see our guide to joint PI and postdoc letters for PhD study. If the student's thesis supervisor cannot write, see PhD references without your thesis adviser.

Before you submit the letter

1. The first paragraph gives your role, the dates and how often you met.
2. The student's own decisions are separated from yours and the group's.
3. One paragraph shows a decision the student made alone: a setback fixed, or a method chosen.
4. Every comparison names a group, a window and a size, and matches any form tick.
5. Co-authored work names who did what.
6. Any limit you mention is explained.
7. The endorsement names the program or research area and uses words a committee can quote.

For the wider picture, read what PhD committees want from letters of recommendation and the master's version of this page, the LOR format and sample for MS and PhD recommenders. If you write from a university whose grades or ranks the reader may not know, our sibling page on explaining grades and rank in a LOR has sentence patterns.

Questions about PhD recommendation letters

How long should a PhD recommendation letter from a supervisor be?

None of the six pages quoted above sets a length for the letter. The sample here is 595 words because a thesis supervisor has five things to say. Use the length the evidence needs, and check the program's form for any limit.

Is a master's thesis supervisor the best PhD referee?

Our view: for an applicant with a research master's, usually yes, because the thesis supervisor is the person who saw the most research. Toronto ECE goes further for its own MASc and PhD: "Applicants who have undertaken a Master's program must include their Master's research supervisor as a referee."

Can the same letter go to US and UK programs?

The core can. For UK programs, add a paragraph on the proposed research area, as Warwick suggests. For Australian programs on a form like Sydney's, move the content into the form's boxes.

What if the student made no mistakes?

Do not invent a setback. Describe the hardest choice the student made without you, and what happened. Writing for a UK PhD? The UK academic reference format covers Cambridge's headings and Warwick's research-capacity ask.

For Hong Kong fellowships, see HKPFS referee reports.

For a master's applicant, the MS CS recommendation letter samples show a manager and a team-project guide.

Get a free review of the letter

Get a free Quick Review of your recommendation letter against GradPilot's graduate recommendation letter rubric, written for faculty and supervisors recommending MS and PhD applicants. It reads for how you know the candidate, incidents rather than adjectives, comparisons that name their group and window, the case for graduate potential, and the strength of your endorsement. For a letter going to programs outside the student's home country, choose the international academic recommendation rubric.

GradPilot's letter reviews aren't for a letter you drafted for a recommender to submit as their own.

More recommender guides:

• LOR from a research internship supervisor: the shorter contact, written for the same PhD reader.
• PhD essays hub: all our doctoral application guides.

Sources

Retrieved October 31, 2026.

• UIUC Siebel School of Computing and Data Science, Applications process and requirements (no cycle stated): letters for PhD applicants address research abilities and potential.
• Carnegie Mellon School of Computer Science, Graduate admissions (no cycle stated): independent research.
• University of Toronto Computer Science, How to apply (Fall 2027): two letters on independent research.
• University of Toronto ECE, Graduate application (no cycle stated): master's supervisor required as referee.
• Stanford Graduate Admissions, Recommendations (no cycle stated): letter prompt and comparison questions.
• University of Sydney, Postgraduate Research Referee Form (form version "vMay16"): capacity question, comparison, scales.
• University of Warwick, Guide to providing academic references (no cycle stated): proposal area and independent research.