The study reports on a validation test of the CROPGRO-Peanut model (Boote et al., 1998) and its subsequent application to assess the yield variability of groundnut in three farming zones of Ghana. The study benefited from previous model calibration and field trial data obtained by Olesen et al in 2001 and 2002. The data obtained by Olesen et al in 2002 as well as those obtained from this study were used for validation and further testing of the groundnut model. The model was evaluated for its performance in simulating the response of two groundnut varieties (Kpedevi, short duration and Goronga, long duration), to planting dates {29th April 2002 and 9th September 2002} and planting densities (D) {9 and 17 plants m-2}. The model predicted the days to 50% emergence, 50% flowering, 50% pegging and 50% pod formation within ±5 days of the observed values. Changes in leaf area index (LAI), and total dry matter significantly correlated with observed values [(R2 = 0.81, and 0.97, respectively, for Kpedevi) and (R2 = 0.86 and R2 = 0.98, respectively, for Goronga)]. The model also accurately simulated the differences in crop growth, yields at final harvest for the different varieties, densities and seasons. The results indicate that under biotic stress-free situations, the model can be used to predict groundnut growth and yields of both varieties as influenced by plant population and planting date. In an application study, the model was run to simulate groundnut yields over 30 years at 3 farming zones, namely, Legon, Kpeve and Akatsi. The generated yield data were ranked from the lowest to highest and were transformed into cumulative relative frequencies {also called Cumulative distribution functions (CDFs)}. The CDFs were obtained for 3 planting dates (early = S1, intermediate = S2, and late planting = S3) as well as 3 planting densities (high = D1, intermediate = D2, and low = D3). Pair-wise comparison of the CDFs indicated that (1): Early planting of both cultivars (Julian date 109 or 19th April), resulted in a lower spread of yield distribution, higher yield and hence least chance of crop failure at Kpeve and Legon. In Akatsi, however, S2 was the preferred planting date for Goronga, although variability of yield data was also very high (CV > 32%). The overall comparison of the yield performance of groundnut at these planting dates followed the trend S1 > S2 > S3. (2):. The general yield ranking of the strategies in which density was varied were D1 > D2 > D3 (i.e. 17 plants m-2, > 13 plants m-2 > 9 plants m-2). The only exception occurred with Goronga cultivar at Akatsi where the preferred density was D2 when sowing was done at S1 (i.e. Julian date 109). (3): Except at Legon, Kpedevi yielded higher at Kpeve and Akatsi than Goronga, irrespective of the planting date and density. Kpedevi cultivar is therefore recommended for these farming zones. At Legon, however, the preferred cultivar was Goronga. In this case seed yield never fell below 1141 kg ha-1. (4): Of the three sites, when assessment criteria excluded other economic analysis, and only yield performance was used to assess preference for a particular strategy, Kpeve was found to be the most suitable site for cultivating Kpedevi, whereas Legon was the preferred site for Goronga. CROPGRO-Peanut model, therefore, presents us with a decision-making tool, which can help design practices that would improve groundnut production in Ghana.