It was shown previously that phage 21 and the defective element e14 integrate at the same site within the icd gene of Escherichia coli K-12 but that 21 integrase and excisionase excise e14 in vivo very infrequently compared to excision of 21. We show here that the reverse is also true: e14 excises itself much better than it excises an adjacent 21 prophage. In vitro integrase assays with various attP substrates delimit the minimal attP site as somewhere between 366 and 418 bp, where the outer limits would include the outermost repeated dodecamers suggested as arm recognition sites by S. J. Schneider (Ph.D. dissertation, Stanford University, Stanford, Calif., 1992). We speculate that the reason 21 attP is larger than lambda attP (240 bp) is because it must include a 209-bp sequence homologous to the 3' end of the icd transcript in order to allow icd expression in lysogens. Alteration of portions of 21 attP to their e14 counterparts shows that 21 requires both the arm site and core site sequences of 21 but that replacements by e14 sequences function in some positions. Consistent with Schneider's in vivo results, and like all other known integrases from lambdoid phages, 21 requires integration host factor for activity.