The present study was conducted in maize (Zea mays L.) on crosses among 20 diverse parents. The materials were evaluated in four different environments for eight characters. Combining ability analysis was carried out following diallel and partial diallel crosses. The number of crosses per parent (s) varied from 3 to 19 and the results were studied to identify the critical value of s that would provide an adequate information with minimum resources. The S5 partial diallel was as good as the S19 for the detection of differences among general combining ability (GCA) effects. Even the S3 gave adequate information in the case of characters with high heritability. However, partial diallel analysis was less efficient in detecting the differences due to specific combining ability (SCA) effects. These results varied with environments, and characters with low heritability were more prone to misinterpretation. GCA effects showed fluctuations in partial diallel analysis which were more pronounced in S5 and S3, particularly for characters with low heritability. The average standard error of difference between GCA effects increased with a decrease in s, with a steep slope for s < 7. The partial diallel analysis was more efficient for the estimation of the variance component of GCA than for SCA, as the estimate of SCA was biased upwards. Estimates of broad sense heritability obtained from the partial diallels agreed with the full diallel analysis better than the narrow sense estimates. Smaller partial diallels gave erratic estimates of heritability, particularly for the characters with low heritability.