Our approach to the study of how the molecular nature of DNA modulates the behavior of mutational sites involves the characterisation of distributions of mutations. The Escherichia coli lacI genetic/M13 cloning system allows the comparison of base substitution frequencies at a large number of sites. The observed distribution of N-methyl-N'-nitro-N-nitrosoguanidine (MNNG)-induced G:C----A:T transition (the predominant event), and A:T----G:C transition (a relatively rare event), is strikingly non-random. Some sites of G:C----A:T mutation are almost 100 times more often mutated by MNNG than the least susceptible sites. Sites of mutation, however, do not display a continuum of mutability, but rather can be strictly demarcated by their 5' flanking base. Sites with a high frequency of occurrence share a common sequence motif, namely 5'-R-G-N-3', which is the sole apparent feature that distinguishes them from sites less commonly mutated (i.e. 5'-Y-G-N-3'). A corollary of this defined site specificity is the absence of a strand bias in MNNG-induced lacI-d mutation. The availability of specific or non-specific alkylation-repair systems does not appear to alter the distribution of mutation, which suggests that the observed mutational distribution is a direct reflection of the initial damage distribution. MNNG does not belong to that class of compounds typified by ultraviolet light or 4-nitroquinoline-N-oxide which exhibit both random and non-random components of mutagenesis.