Two basic types of models, conservative and replicative, have been proposed to account for the mechanism of transposition in bacteria. A method was developed to test these models by positive selection of various transposon-promoted events as galactose-resistant colonies from plasmid-containing cells. The results show that recA plays an important role in the transposition of Tn5 and Tn9 in Escherichia coli. All Tn5-promoted events (cointegrates, deletions and transpositions) are suppressed in recA-, and restored in recA+. In the case of Tn9, however, only transpositions (but not cointegrates or deletions) are diminished in recA-. Therefore, the recA function is required for cointegrate formation by Tn5, and for cointegrate resolution by Tn9. Both Tn5 and Tn9 cointegrates segregate transpositions (which can be seen as sectors on indicator plates) in recA+ hosts. In recA-, the unresolved Tn9 cointegrates undergo a second round of cointegrate formation to excise plasmids bearing galactose-resistant deletions. In growing cultures, the proportion of cointegrates declines steadily while transpositions increase so that, in late stages, cultures are rich in transpositions and contain few cointegrates. This explains the failure of previous workers to identify cointegrates as essential intermediates in transposition. Hence, with the exception of the recA requirement, the mechanism of transposition of these composite transposons is not very different from simple transposons like Tn3. It is concluded that transposition of Tn5 and Tn9 is normally a replicative process.