We have presented the results of experiments with IS903- and IS10- derived transposons that have led us to the following conclusions: The predominant mechanism of transpositional recombination of these IS elements is a donor-suicide process that results intermolecularly in a simple IS insertion. This process presumably involves little or no replication of the IS. Intramolecular transposition by this process normally results in nonviable products. However, in the particular situation where the transpositional target lies within the transposon, viable products are obtained; these are deletions and deletion-inversions. Deletions between an IS and a target lying outside the element (the conventional "adjacent deletion") occur by a fully replicative process analogous to the formation of cointegrate molecules in intermolecular transposition. The ability of an IS to promote adjacent deletions correlates closely with its ability to fuse replicons into a cointegrate. Before transposition can occur, a complex of the transposase and both IS ends is probably formed. Requirement for such a pretranspositional complex is suggested by the effect on transpositional frequency of changing the distance between the ends. Our results do not support any of the asymmetrical models for transposition. They are, however, compatible with a modified version of the symmetric model proposed by Shapiro (1979). It is interesting to note the similarity between the structures generated by intramolecular simple transposition of an inverse transposon and the circular structures apparently formed by retroviral and copia autointegrative transposition. Shoemaker et al. (1981a,b) and Flavell and Ish-Horowicz (1983) have characterized circular molecules from retrovirally infected cells and Drosophila tissue-culture cells, respectively. The structures of some of the circular molecules resemble deletions and deletion-inversions (Fig. 3B). To our knowledge, a circular species containing two long terminal repeats (LTRs) and an adjacent deletion, which we predict could only occur by a fully replicative process given the similarity in geometry of an LTR to an IS, have not been found. It would appear, then, that the molecule containing two LTRs acts as an inverse transposon, integrating into itself. Shoemaker et al. (1981b) and Flavell and Ish-Horowicz (1983) have also suggested that these products arise from molecules containing two LTRs. We suggest that the two inside LTR ends interact in a conservative, intramolecular, simple transpositionlike event.