"How many themes should I have?" is one of the most common questions from people writing up a thematic analysis, and the honest answer is disappointing: there is no right number. What exists instead is a set of signs that your number is wrong, and a way of deciding what belongs together. This article walks through both, so you can defend whatever number you end up with.
Why no number is correct
In the reflexive version of thematic analysis developed by Virginia Braun and Victoria Clarke, themes are not counted out of the data like items in a bucket. A theme is a pattern of shared meaning built around a central idea, and you construct it through interpretation. A different researcher, or the same one with a different question, could build a different set. Braun and Clarke have also argued against treating a fixed point of "saturation" as the signal that analysis is done, because themes are generated rather than found, and the same logic applies to their number. If you work in a codebook or coding-reliability tradition, the frame is fixed earlier and the count is easier to defend, but even there nobody can say in advance that seven is right and twelve is wrong. Our article on reflexive versus codebook thematic analysis explains why this difference matters for everything else you report.
So the useful question is not how many, but whether your themes do their job: answer the research question, hold together internally, and stay distinct from each other.
Signs you have too many
- Themes that are really topics. A label such as "workload" or "communication" names an area of the data, not a claim about it. A long list of such labels usually means you are summarising, not analysing.
- Overlap you keep apologising for. If you need sentences like "this is related to theme 4" in nearly every section, two themes probably share one central idea.
- Thin themes. A theme supported by one or two extracts from a single participant is rarely a pattern. It may be a good observation, but it belongs inside a larger theme or in a sentence of its own.
- No room to write. If your results section allows roughly a page per theme and you have fifteen themes, each will get a paragraph and a quote. Depth is what readers remember.
Signs you have too few
- A theme that is a container. If one theme collects half your extracts and you cannot state its central idea in a sentence, it probably hides two or three themes under one broad name.
- Internal contradiction. When extracts under one theme point in opposite directions without you saying so, either the contradiction is the finding or the theme needs splitting.
- Unanswered parts of the research question. Read the question again. If one part has no theme that speaks to it, you are either missing a theme or the question was broader than your data.
Typical mistake: counting themes to look rigorous
A common mistake is to treat the number itself as evidence of quality, either by adding themes to look thorough or by merging them to look tidy. A related one is to report how many participants mentioned each theme as if it showed importance. In reflexive work, frequency does not make a theme; meaning does. In codebook work, reporting how many responses fall under each category is legitimate, but then the categories have to be defined clearly enough that the counts mean something. Decide which of the two you are doing before you start counting.
Themes and subthemes: what the levels are for
Subthemes are not a way to keep many themes while appearing to have few. Use them when a theme has one central idea that shows up in distinguishable ways, for example the same underlying tension expressed by two kinds of participants. If you cannot say what unites the subthemes, they are separate themes. A thematic map with a few themes and a handful of subthemes is usually easier to read than a flat list, but only if the grouping reflects a real relationship in the data. The phase on reviewing and defining themes in our walkthrough of the six phases of thematic analysis shows where this decision belongs in the process.
A practical test before you finalise
For each candidate theme, write two things: a one-sentence statement of its central idea, and a sentence saying how it differs from the nearest other theme. If you cannot write the first, the theme is a topic. If you cannot write the second, merge or rethink. Then check that the set as a whole tells a story that answers your question. If you are still unsure how to get from codes to this stage, the guide to coding qualitative data covers the earlier step of building the codes themes are made from.
What software can and cannot do here
Software can show you what is in the material. In Themera, every code is backed by a verbatim quote, and quotes that are not exactly in the material are rejected automatically, so you can check whether a candidate theme really rests on several voices rather than one. A coverage indicator shows which responses received no code, which helps you see whether your themes leave part of the material unexplained, and you can edit the codebook and re-code the material when you merge or split. What no tool can do is decide that two patterns share a central idea. That judgement is the analysis, and it stays with you.
A formulation for your methods section
Adapt this to what you actually did. It is a template, not a citation.
Themes were developed iteratively. Initial codes were grouped into candidate themes, which were reviewed against the coded extracts and against the full dataset. Candidate themes were merged when they shared a central organising idea and split when they contained distinct ideas. The number of themes was not set in advance; the final set of [N] themes and [M] subthemes was chosen because each theme can be stated as a single central idea, is supported by extracts from multiple participants, and together the themes address the research question. Computational support was used for a first coding pass; all codes and themes were reviewed and revised by the researcher.
If your supervisor or target journal expects a particular range, ask them directly. That is a question about local convention, and they are the right people to answer it.
In short
Stop looking for the right number and start testing each theme: one central idea, distinct from its neighbours, grounded in several extracts. Then say in the methods section how you decided.
See a worked example analysis with codebook, verbatim quotes and coverage, no account needed, or try it on your own data: three analyses free, no card needed.