Correlation and path analyses between yield components and grain yield may be useful to provide the basis for planning of more efficient in selection program. The objectives of the experiment were to investigate the most responsible yield components to grain yield of morphologically diverse soybean varieties. Twelve soybean varieties were evaluated on rainfed wetland after rice in Grobogan, Central Java, during the late dry season of 2014. The experiment was arranged in randomized block design with four replications. Each of the genotype was grown in five rows of four and half meters long. Plant spacing was 40 cm x 15 cm, with two seeds per hill. Results showed that among the yield components, number of pods per plant, seed size and harvest index were each positively correlated with grain yield (r = 0.530; 0.376 and 0.608). The direct effect of pods per plant, seed size and harvest index to grain yield as indicated by path coefficient were among the highest, each was 0.813, 0.540 and 0.901 respectively. The direct effects of other yield components were small or negative. Based on both analyses, soybean genotypes with high grain yield should have a high number of pods per plant (>40 pods), large seed size (>15 g/100 seeds) and high harvest index (>50%). Therefore, pods per plant, seed size and harvest index can be used as selection criteria for developing varieties for rainfed wetland. Although the direct effect of days to maturity to grain yield was small, but early maturing varieties should be considered as selection criterion, because on rainfed wetland during late dry season the availability of soil moisture often is uncertain. Early maturing varieties can be more useful under this condition to minimize yield losses due to drought stress.