Early potassium fertilization in previous crops may be beneficial to the main crop in succession. The objective of this study was to investigate the behavior of the soybean crop as a function of potassium (K) doses in the previous crops, associated or not with K application in soybean. The experiment was carried out from 2012 to 2015 in an experimental area of the Faculty of Engineering, UNESP, Campus of Ilha Solteira-SP, located in Selvíria-MS, Brazil. For the previous crops, a randomized block design was used in a 3 x 4 factorial scheme with four replications, i.e., three previous crops (maize, sorghum and millet) and four K doses (0, 35, 70 and 120 kg ha-1 of K2O). For soybean in succession, a randomized complete block design was used in the split-plot scheme. The treatments were arranged in a 3 x 4 x 2 factorial scheme with four replicates, i.e., soybean sowing on three types of straw (maize, sorghum and millet), residual of the four K doses and with or without top-dressing K fertilization in the soybean. Millet accumulated higher K content in comparison to maize and sorghum. The return of K to the system by millet is similar to that by maize. Potassium doses in the previous crops do not alter their dry matter production and K content. Potassium fertilization in the soybean crop in succession to millet allows higher yields.