Abstract: Two sets of test materials were derived from diallel cross among nine different types of selected rice varieties (lines). Based on the experimental data at different development stages of early and late rice, the estimation of heterosis and genetic correlation of rice grain weight at some development stages (0^→ t) or at some special development phases (t-1^→t) was conducted by using additive dominant genetic model and developmental genetic model for quantitative traits. The results indicated that the heterosis over mid parent value based on population mean was not significant at 5% level at the early stages (1～12 d), and turned to positively significant as the development process proceeded in early rice. The heterobeltiosis based on population mean appeared to be small negative during the entire courses of grain filling. The heterosis over mid parent based on population mean appeared to be positively significant at the early and middle grain filling stages (1～18 d), but became insignificant at the late stages (19～28 d) in late rice. The heterobeltiosis based on population mean appeared to be positively significant at the early grain filling stages (1～12 d), but turned to be significantly negative as the development process proceeded in late rice. It also indicated that the heterosis of grain weight development was beneficial to improve the quality of grain filling in late rice, compared to that in early rice. The genetic correlation between the grain weights per se at different stages and the final grain weight was mainly controlled by dominant effects, and became closer with the development process in the early rice, while by additive effects in the late rice. It's suggested that indirect selection of final grain weight throughout the process of grain filling would be more effective in late rice.