Crop models can support agricultural decisions, yet their reliability is necessarily limited when they do not sufficiently represent the complexity and specific circumstances of the target system. In some cases, models have such prohibitively high data requirements that they are only applicable with far-reaching and often questionable assumptions. In this paper, we demonstrate a customizable solution-oriented approach for crop modelling in situations where data and resources are limited. To address system complexity and produce a probabilistic crop model that does not depend on precise data, we used participatory analysis to describe system components using individual Bayesian networks that formalize expert knowledge into probabilistic causal relationships among important variables. We then used these Bayesian networks to generate inputs for a Monte Carlo model that illustrates the determinants of crop growth and simulates plausible ranges of expected grain and biomass yields at various stages of crop development. The resulting model accounts for all important variables and their in teractions, as examined by local and foreign experts and described in relevant literature. We describe how to develop and customize such a model to specific situations based on case studies related to flood-based farming systems in Ethiopia and Kenya. The model assesses the performance of cropping systems and individual crops, and identifies factors of high importance for system outcomes. This approach to crop modelling paves the way for new opportunities to support agricultural decisions, since it does not require perfect information and can accommodate system complexity and uncertainty in data-poor environments.