The ink drop spread (IDS) method is a modeling technique that has several advantages in robustness, real-time capabilities, tractability, and interpretability of models; thus, it has a good potential to be a useful soft computing tool. In this method, the structure of models is determined by the partitioning of the input domain; finding the optimal number of partitions is the most effective means for achieving high model accuracy. This paper proposes a basic constructive algorithm for the structural optimization of IDS models and presents the performance of the IDS modeling in regression and classification tasks using three-input nonlinear systems.