The automatic design of fuzzy rule-based models and classifiers from data is considered. It is recognized that both accuracy and transparency are of major importance and we seek to keep the rule-based models small and comprehensible. An iterative approach for developing such fuzzy rule-based models is proposed. First, an initial model is derived from the data. Subsequently, a real-coded genetic algorithm (GA) is applied in an iterative fashion together with a rule base simplification algorithm in order to optimize and simplify the model, respectively. The proposed modeling approach is demonstrated for a system identification and a classification problem. Results are compared to other approaches in the literature. The proposed modeling approach gives more compact, interpretable and accurate models.