Soil sampling approaches have been compared based on soil-test variation. This study evaluated sampling approaches for P and K based on yield response to fertilization. Strip trials were established on four fields for P and three fields for K managed with corn (Zea mays L.) and soybean (Glycine max L. Merr.) rotations and evaluated 3 or 4 yr (27 site-years). Treatments replicated three to four times were fertilizer and no fertilizer application. Soil test results from a dense grid point sampling (DG) approach (0.08 to 0.27 ha) were used to simulate six approaches: (i) 1.0-ha grid cells (GC), and zones delineated based on (ii) soil series from digitized survey maps (SMZ); (iii) elevation (EZ); (iv) apparent soil electrical conductivity, ECa (ECZ); (v) EZ and ECZ (EECZ); and (vi) EZ, ECZ, and slope (EECSZ). Grain yield monitors, global positioning systems (GPS), and geographical information systems (GIS) were used to describe crop responses. Estimates of soil-test variation were largest for DG, intermediate for GC, and less for other approaches. Crops responded (P # 0.05) to fertilization in 20 site-years. Sampling approaches DG, GC, EZ, EECZ or EECSZ, ECZ, and SMZ identified a differential within-field yield response in 16, 8, 5, 3, 2, and 2 site-years, respectively. Differential yield responses seldom were explained by zone-mean soil-test values. Zone approaches often identified areas with different yield levels but were less effective than DG or GC at describing within-field variation of soil tests and yield response to fertilization. Zone approaches may be more effective in fields with shorter fertilization histories or soils with more contrast in properties.