The leucine zipper is a dimerization domain occurring mostly in regula
tory and thus in many oncogenic proteins. The leucine repeat in the se
quence has been traditionally used for identification, however with po
or reliability. The coiled coil structure of a leucine zipper is requi
red for dimerization and can be predicted with reasonable accuracy by
existing algorithms. We exploit this fact for identification of leucin
e zippers from sequence alone. We present a program, 2ZIP, which combi
nes a standard coiled coil prediction algorithm with an approximate se
arch for the characteristic leucine repeat. No further information fro
m homologues is required for prediction. This approach improves signif
icantly over existing methods, especially in that the coiled coil pred
iction turns out to be highly informative and avoids large numbers of
false positives. Many problems in predicting zippers or assessing pred
iction results stem from wrong sequence annotations in the database.