I have condensed and re-ordered these points from An Experiment with Time by J.W Dunne, Revised third edition of 1934 - Chapter III - The Experiment.
1. The aim is to make a detailed written record of dreams which, when completed, can be examined to distinguish memories of past events, but more importantly can be examined over the following days/ weeks/ months to discover whether the dream predicted future experiences (whether real life experiences or discoveries made in news and media reports, books, movies etc).
2. A notebook and pencil are kept under the pillow and immediately upon waking, and before opening eyes the experimenter should try to recall the dream which is rapidly vanishing from memory.
3. Until the noting is finished do not do anything else, do not talk, do not think of anything else except the dreams.
4. This may be a single incident, or a mood - this should be fixed on and you should try to remember the details. If this is impossible, at least record what-you-are-thinking-about at the moment of waking.
5. There may then emerge a section of dream, or further isolated incidents from previous dreams. In noting dreams, the experimenter should focus on facts, images, events, appearances, sensory impressions etc rather than on interpretations of meanings, explanations, causes etc.
6. These should be briefly noted in the book – a word or two for each. Note as many details as possible – especially when details are unusual or happen in unusual combinations.
7. Then go through the incidents one at a time starting with the first recollection – concentrate until further detail or story is recalled and note it. Do the same for the other incidents.
8. Read through what you have, memorise, try to re-visualise each aspect of each detail or story fragment.
9. At the end of each day, read your dream records through and see if any were predictive of any aspects of what happened.
10. The principle is that constellations of more details and of shorter time limits in the future will make coincidence less probable – the more specific details of a dream situation that are replicated over a shorter future time span, then the less likely that this could be due to random coincidence.
11. Mundane dream events are more likely to happen by chance in a period of mundane life – and unusual dream contents are less likely to happen by coincidence.
12. Therefore the best time for the dream experiment is before a novel or unpredictable experience such as a journey, a holiday or some other break in monotony.
13. Each of the noted dreams should be marked with a cross for a decisive prediction of an exceptional replication (e.g. unusual dream and unusual experience to follow), and a circle for probable but less certain replications. A few crosses are equivalent to a lot more circles, but both might be convincing if enough accumulate.
14. Anticipate that the waking mind will resist associations between a dream and subsequent event – therefore read the dream records with care. Associations between dreams and the past will be obvious and acceptable to the mind as obviously causal; but there is an inbuilt reluctance to recognize associations with the future – to do this is more like a process of pattern recognition, and the experimenter tends to become distracted by stories and meanings. Even apparently trivial or tenuous associations need to be properly followed-up and evaluated.