Maize produced under rainfed conditions is a common practice in Brazil. Crops under this production system are highly affected by the inherent annual and intra-seasonal weather variability, especially dry spells, that affect productivity. One of the simplest strategies with virtually no cost to mitigate this problem is the determination of a sowing window that minimizes the negative effects of weather variability on rainfed agriculture. The objectives of this study were to: a) use the results of maize yield simulated with a process-based model to establish sowing windows and, b) compare our results with the current methodology employed by the Brazilian Ministry of Agriculture (MAPA). The CSM-CERES-Maize model of the Decision Support System for Agrotechnology Transfer (DSSAT) was used to simulate scenarios of weekly sowing dates, under rainfed conditions, for selected counties of Minas Gerais State, Brazil. For each sowing date it was determined the yield break by comparing the average yield of the current sowing date with the highest average yield obtained from all sowing dates. Thus, each sowing date was associated with a risk of yield loss. The use of simulated yield break data to establish sowing windows produced consistent results. The sowing windows obtained from the simulations were slightly different from those proposed by MAPA. The use of process-based models to simulate crop yield allows for the integration of many factors not considered in the current crop zoning approach used by MAPA. The proposed approach has advantages over the MAPA methodology in that it includes the possibility of determining the expected average yield and its amplitude.