Methods that estimate the probability of agricultural drought using atmospheric data can be widely applied but have not been compared with actual drought occurrence indicated by soil moisture measurements. Our objectives were to develop a drought probability assessment method using long‐term measurements of soil water deficits (SWDs) and to compare the resulting probabilities with those of an existing method based on atmospheric water deficits (AWDs). Fifteen years of daily precipitation, air temperature, and soil moisture measurements for eight locations across Oklahoma were used to calculate the probability (P) of water deficits sufficient to cause plant water stress for each day of the growing season. For the SWD method, the drought threshold was set at 50% depletion of the soil’s total available water capacity. For the AWD method, the threshold was a 7‐d cumulative AWD of 50 mm. Values of AWD were similar to calculated SWD for the 0‐ to 40‐cm soil layer in the spring; however, AWD values seldom reached the drought threshold. Thus, drought P values calculated by the AWD method were unreasonably low and consistently lower than P estimated by the SWD method. The AWD method showed greater agreement with the SWD method when 37 mm was used as the AWD threshold or when the original 50‐mm threshold was applied for a 15‐d cumulative AWD. The new SWD method gave plausible and consistent results when applied to both the 0‐ to 40‐ and 0‐ to 80‐cm soil layers and should be utilized when long‐term soil moisture data are available.