Increasing grain weight is one means of increasing rice (Oryza sativa L.) grain yields. Selection for increased density of filled grains may offer an approach to increasing rice grain weight. Before rice breeding programs can begin effectively selecting for higher grain density, the nature and amount of the genetic variation present must be evaluated. A Design II mating plant with two sets was constructed using 16 parents. The 16 parents were representative of cultivars and elite breeding material available to breeders of long-grain rice in the southern U.S. The parents and 32 F1 hybrids were evaluated in 1990 at two Arkansas locations: Stuttgart and Marianna. Additive variation was essentially zero for grain density as indicated by nonsignificant general combining ability (GCA) and specific combining ability (SCA) mean squares and a narrowsense heritability estimate −0.04 (±0.07). Increasing rice grain weight through increased grain density would not be feasible in U.S. southern long-grain rice unless new germplasm with higher filled grain densities is introduced. Genetic variation for grain weight was predominately additive and complemented with additive x additive epistatic variation, indicating that selection for increased grain weight could be practiced in segregating generations. The small proportion of epistatic variation to additive variation would not justify delaying selection until epistatic combinations are fixed in the homozygous line. Midparent grain weight means would serve as an adequate indicator of progeny performance for cross appraisal.