Geographical patterns of mtDNA variation were studied in 12 Italian samples (1072 individuals) by two different spatial autocorrelation methods. Separate analyses of the frequencies of 12 restriction morphs show North-South clines, differences between Sardinia and the mainland populations, and the effects of isolation by distance. A recently developed autocorrelation statistic summarizing molecular similarity at all sites (AIDA; autocorrelation index for DNA analysis) confirms the presence of a clinal pattern; differences between random pairs of haplotypes tend to increase with their geographical distance. The partition of gene diversity, however, reveals that most variability occurs within populations, whereas differences between populations are minor (GST = 0.057). When the data from the 12 samples are pooled, two descriptors of genetic variability (number of polymorphic sites and average sequence difference between pairs of individuals) do not behave as expected under neutrality. The presence of clinal patterns, Tajima's tests, and a simulation experiment agree in suggesting that population sizes increased rapidly in Italy and Sicily but not necessarily so in Sardinia. The distribution of pairwise sequence differences in the Italian peninsula (excluding Sardinia) permits a tentative location of the demographic increase between 8000 and 20,500 years ago. These dates are consistent with archaeological estimates of two distinct expansion processes, occurring, respectively, in the Neolithic and after the last glacial maximum in the Paleolithic. Conversely, there is no genetic evidence that such processes have had a major impact on the Sardinian population.Most studies of mtDNA variation in humans have inferred evolutionary processes by reconstructing history-i.e., genealogies of haplotypes (1-4). With one exception (5), geographical information has been digregarded, or it has been used simply to classify populations, as an alternative to the subjective criteria of racial classification (6, 7). There is no doubt, however, that spatial patterns of genetic diversity also contain useful information for evolutionary inferences. One reason for this omission may lie in the relative paucity of statistical tools suited for spatial analysis of molecular data.Genetic variation in space can be summarized by spatial autocorrelation measures (see, e.g., ref. 8). These statistics have also been used to test hypotheses on past demographic processes (9, 10). Here we apply them to a data base of mtDNA restriction fragment length polymorphism data. We reconstruct patterns of genetic variation from molecular information, and we try to draw inferences about the underlying microevolutionary phenomena.In the first part of this paper, we analyze the frequencies of 12 common restriction morphs, treating them as if they were allele frequencies. In the second part, we use a recently developed autocorrelation statistic, AIDA (autocorrelation index for DNA analysis) (11), which compares DNA sequences (and not only their frequencies) ...