Most linkage and population genetic studies that use microsatellites assume that the polymorphism observed at these loci is due simply to variation in the number of units of a single repeat. Variation is far more complex, however, for the numerous microsatellites that contain interruptions within the repeat or contain more than one type of repeat. We observed that for D18S58, a compound microsatellite containing (CG)(m), as well as (CA)(n) repeats, the apparent length of certain alleles varied between genotyping experiments. Similar results were obtained with other (CG)(m)-(CA)(n) repeats. Sequencing demonstrated that the D18S58 alleles demonstrating variable mobility contained longer (CG)(m) stretches than those alleles whose length did not appear to vary between experiments. These results suggest that (CG)(m) repeats, which are frequently present in compound human microsatellites, are prone to form an unusually stable secondary structure. We discuss the relative frequency of different classes of compound microsatellites identified through database searches, as well as their patterns of sequence and variation. Further characterization of such variation is important for elucidating the origin, mutational processes, and structure of these widely used, but incompletely understood, sequences.