Thirty genotypes from Jilin and Liaoning provinces were tested in a four-replicate randomized block design to investigate the relationships (including phenotypic, genotypic and varietal) among twenty-five characters, and to constructindices for yield per plant and yield per plot in soybean. Some theoretical problems relating to genotypic correlation coefficient and selection index were also discussed. The results of this study indicate that yield per plot is positively correlated with yield per plant, apparent economic index, ratio of seed weight to dry weight per plant, dry weight per plant, number of seeds per plant, number of pods per plant, number of pods per node;but negatively correlatod with plant height and protein content. Similar correIation is also found between yield per plant and the traits mentioned above. The path-coefficient analysis of varietal correlation coefficient between yield components suggest that dry weight per plant, number of seeds per plant, 100 Seed weight and ratio of seed weight to dry weight per plant be the most important contributors to yield production. One hundred and seventy selection indices were constructed for yield per plot and yield per plant respectively. The most effective index for yield per plant is composed of number of pods at 10 cm. height, number of nodes per plant, ratio of seed weight to dry weight per plant and 100 seed weight with the relative selection efficiency of 161.65%.100 seed weight is the domonant factor in constructing indices for yield per plant. The index for yield per plot constructed by number of nodes per plant, number of pods per plant, number of seeds per pod, yield per plant and yield per plot has the highest relative selection efficiency (111.7%). in an effective selection index, yield per plot itself must be included. Considering the unavailability of the methods for testing genotypic correlation coefficient, it is suggested that varietal correlation coefficient be used as a substitute for genotypic correlation coefficient in interpreting the relationships among quantitative characters and for other purposes such as path-coefficient analysis.