Variable numbers of tandem repeats are valuable markers in genetic studies. The arrays of interest are simple microsatellites, containing repetitions of 1-5 nucleotides, and minisatellites, with multiple iterations of approximately 10 to 100 bp. Microsatellite origins can be explained by replication errors in regions fortuitously containing two or more adjacent short repeats. Microsatellite variation arises by replication errors in the absence of mismatch correction (R. Parsons et al., 1993, Cell 75: 1227-1236; M. Strand et al., 1993, Nature 365: 274-276). Variation in the size of minisatellites is thought to involve homologous recombination processes, including gene conversions (J. Buard and G. Vergnaud, 1994, EMBO J. 13: 3203-3210; A. J. Jeffreys et al., 1994, Nature Genet. 6: 136-145) and possibly unequal exchanges among repeats. However, the origins of minisatellites are less obvious. The probability of finding a direct tandem repeat of minisatellite size by chance alone is very low [<4-(10 to 100)]. Here we report the finding of short direct repeats of 5 to 10 bp flanking many yeast and human minisatellites that may be involved in their origins through replication slippage or unequal crossings over.