The influence of sequence context on the ability of DNA polymerase to bypass sites of base loss was addressed using an in vitro selection system. Oligonucleotides containing either an aldehydic abasic site or tetrahydrofuran surrounded by four randomized bases on both the 5' and 3' sides were used as templates for synthesis by phage T4 DNA polymerase holoenzyme proficient or deficient in the 3'-->5' proofreading exonuclease activity. Successful bypass products were purified, subcloned and the sequences of approximately 100 subclones were determined for each of the four polymerase/lesion combinations tested. Between 7 and 19 % of the bypass products contained deletions of one to three nucleotides in the randomized region. In bypass products not containing deletions, biases for and against certain nucleotides were readily noticeable across the entire randomized region. Template strands from successful bypass products of abasic sites had a high frequency of T in most of the randomized positions, while those from bypass products of tetrahydrofuran had a high frequency of G at the positions immediately to the 3' and 5' side of the lesion. Consensus sequences were shared by successful bypass products of the same lesion but not between bypass products of the two lesions. The consensus sequence for efficient bypass of tetrahydrofuran was over-represented in several frames relative to the lesion. T4 DNA polymerase inserted A opposite abasic sites 63 % of the time in the presence of proofreading and 79 % of the time in its absence, followed by G>T>C, while the insertion of A opposite tetrahydrofuran ranged between 93 % and 100 % in the presence and absence of proofreading, respectively. Finally, sequence context influenced the choice of nucleotide inserted opposite abasic sites and consensus sequences which favored the incorporation of nucleotides other than A were defined.