Restriction fragment length polymorphisms (RFLPs) have been proposed for investigating relationships among inbred lines and predicting heterosis and performance of single-cross hybrids in maize (Zea mays L.). Such use was evaluated in 20 maize inbreds classified as first-cycle, second-cycle, good, and poor lines, and in diallel crosses within types. Eight generations (parents, F₁, F₂, F₃, backcrosses, and backcrosses selfed) from 67 crosses were evaluated for grain yield in five Iowa environments. Genetic effects were estimated from generation means by ordinary diallel analyses and the Eberhart-Gardner genetic model. Poor inbreds showed significantly greater average heterosis than did good lines. Estimates of additive ✕ additive epistatic effects were negative and highly significant in all except first-cycle lines. Using two restriction enzymes and 82 genomic) DNA probes distributed throughout the maize genome, all but one probe revealed polymorphisms with at least one enzyme (~4.5 variants per RFLP locus). Genetic distances between lines within types were estimated as Rogers' distances (RD). Within diallel sets, RD values were partitioned into general (GRD)and specific (SRD). All four types of lines showed similar means and substantial variation for RD; GRD explained ~40% of the variation among RD values. Cluster analyses revealed associations among lines generally consistent with expectations based on known pedigrees. Correlations of RD and SRD with F₁ yield heterosis, specific heterosis, and specific combining ability were positive but small (r = ≤ 0.50) when combined for all crosses. Results indicated that RFLPs can be used to investigate pedigree relationships among maize inbreds, but also suggest that RFLP-based genetic distance measures are of limited use in predicting heterotic performance of single crosses between unrelated lines.