First, chronic administration of melatonin at .2 mg or above will create a very interesting biological state in the brain. The natural release of melatonin peaks at 0.2 mg at the same time as growth hormone, cortisol & slow-wave sleep. It is not associated with REM sleep. It's more likely that increased dream recall is occuring because of a slight REM rebound at the end of the night, a compensatory reponse to an increase in NREM, especially slow-wave sleep in the beginning of the night. Tryptophan is also associated with deep NREM sleep, not REM. Tryptophan depletion results in shorter REM latency and maybe increased REM time. Moreover, both melatonin & tryptophan are related to serotonin production in the brain. Increasing serotonin, or using drugs that inhibit serotonin re-uptake, such as Prozac, Zoloft & Paxil, leaves more serotonin available in the brain and should, theoretically, reduce REM sleep, not increase it. (That's exactly what these drugs do, but they are also alerting & produce sleep fragmentation & microarousals, particularly at the end of the night. Lightening or fragmenting sleep should improve the likelhood of recalling a dream). The natural physiological level of melatonin is a hundred-fold smaller than what is available commercially. For every day use, 0.2 mg is much more appropriate. To get over shift work, jet lag, or in some individuals with biological rhythm disturbances, you may need a higher dose, but 2 mg every day if you do not fall into one of the categories above is overkill and may be associated with headache, and worse sleep disturbance than was evident prior to taking the melatonin. I urge extreme caution with high doses of melatonin because there is NO good research on the long term effects. The research data does not back up the grandiose claims of melatonin as a wonder drug. It does shift the biological clock and can at small doses improve sleep, particularly ease of sleep onset, and it does minimize cardiac stress response in hamsters with chronic heart disease, but that is the sum total of the good data.