A Thesis is:

A thesis is the acquisition and dissemination of new knowledge. In order to demonstrate this the author must demonstrate that they understand what the relevant state of the art is and what the strengths and weaknesses of the SoA are. For someone's work to be knowledge there must be a demonstration that suitable and systematic methods were used to evaluate the chosen hypothesis. It is important that "new" is not just new to the researcher, but also new to the community - PhDs were sometimes in the past failed because a paper was published by another researcher a few weeks previously dealing with the same work. I don't believe this is as common today, but novelty/originality/new understanding/marshalling existing ideas in ways that provide new insights is what it is all about.

A PhD Thesis Must Contain:

Knowledge, understanding, and appreciation of the field This will show motivation, relevance to X, Y, & X, who is doing what, &c. Critical analysis of related work. Person X is doing Y, this is important because ..., this doesn't address these points ... Link the failings of related work to your own work. Importance (relevance) of own work. State contributions, is this an incremental improvement on the state of the art, an evolution on existing work, &c. Beware of appearing to be too original, don't appear to have missed or ignored existing work.

A PhD Thesis is Not:

Not "a diary of work done". In order to be awarded a PhD you must be able to present your work so that it is accessible to others and so that it demonstrates your mastery of a given subject. Although PhD theses may differ widely, you certainly won't be awarded a PhD just for doing three year's work and you won't be awarded a PhD for "a diary of work done". A common attitude is "well, I've done my PhD, now all I've got to do is write it up". Beware! The thesis IS the PhD - it doesn't really matter how great your research has been during the three years - all that really matters is the thesis. Not "a collection of papers". At UCL this is not an acceptable PhD thesis (some other universities allow this as a PhD route, e.g. for staff, but the required standard is very high). At UCL your thesis must have a THEME. It is similar to writing a book. You can however take a collection of papers and turn it into the core of a PhD. Not "a big 3rd year project". Though some 3rdyr projects are excellent, most do not contain sufficient critical analysis or scientific method. Not "a lone journey". It is important to have other people involved, if for nothing else then for proofreading. You need to have an experienced supervisor who can tell you when to stop! (this is often the biggest problem faced by students). As the person doing the PhD, you are too involved and therefore you have the worst judgement on what is good or bad - you must get external advice. Also remember that a thesis should be designed for the benefit of the reader, not the writer! So get lots of people to read your thesis and tell you what parts they could not understand.

Typical PhD Thesis Layout

Note: Should have an odd number of chapters, between 5 and 9. Abstract 1. Introduction Set the scene and problem statement. Introduce structure of thesis, state contributions (3-5). 2. Background Demonstrate wider appreciation (context). Provide motivation. The problem statement and the motivation state how you want the PhD to be judged - as engineering, scientific method, theory, philosophy, &c. 3. Related Work Survey and critical assessment. Relation to own work. 4-6. Analysis, design, implementation and interpretation of results 7. Critical assessment of own work State hypothesis, and demonstrate precision, thoroughness, contribution, and comparison with closest rival. 8. Further Work 9. Summary Conclusions Restate contribution Appendix Bibliography

Notes
A PhD made up on only critical assessment may be possible (for UCL) but is extremely difficult. Average, good, size for a thesis is 150 pages all in. Perhaps up to 50 extra pages for a big appendix and bibliography. Beware of the trend to write long and boring doctorates (papers, &c), improve your communications skills. Another important datapoint: 2-3 conference, or 1-2 journal papers in respectable (ACM, IEEE, IOP like) places are good enough for chapters 4,5,6, and therefore the core of a PhD - testing by publication is a VERY good defense (or defence). Also note that the feedback from reviewers is extremely helpful, so all PhD students should be trying to publish their work (the feedback is even more useful when your submissions don't get published!).

Always think - Presentation. Be precise in all things, esp: the statement of the problem, the solution, methods and frameworks. Thoroughness == scientific method. You must show proof that your contributions are valid. Chapter headings - use 7 or 9! An odd number of (total) chapters gives a balanced appearance to the work (CC has a reference to back this up).