@Frankie While it is true that Kinsey made his estimate based on behavior, it was not based on one-time behavior. That is to say, a single act of homosexual sex did not get someone classified as homosexual under his system (if it did, he would have reported that 37% of the population was homosexual). That was the whole point of having a scale, after all: it allowed people to be classified on a continuum rather than in binary terms.
The problem regarding self-identification arises with the way that other researchers classify their subjects. Self-identification as heterosexual is a personal choice. But as important as that choice may be for the individual, its scientific relevance is limited. Two people who have had the same number and distribution of partners (say, 2 males and 7 females in the same order and at the same times of their lives) could self-identify differently (one as heterosexual and the other as bisexual, for example). Kinsey’s classification schema is more useful if you are trying to make assessments about behavioral demographics, but the self-identification data is also interesting for various other purposes (especially when correlated with the behavioral data).