This paper presents a novel approach to symbolic music generation using Modal Decision Trees (MDTs), a class of interpretable, rule-based models that integrate modal logic into the decision-tree algorithm. We construct a dataset from multitrack MIDI files, representing each track as a binary pianoroll matrix, and frame the generation task as a binary classification problem: predicting whether a pitch should be played at each time step based on a local window of context. The performed experiments demonstrate that MDTs, despite not leveraging explicit pitch identities, achieve strong performance, with an F1 score of 0.926 and balanced accuracy of 0.943. The generated music closely follows the input melodies, with some modifications that reflect the model’s learned patterns. While MDTs do not yet match the performance of deep learning models for temporal sequence modeling, our results highlight their potential as interpretable tools for symbolic music generation. We discuss future directions, including modal decision forests, multi-track analysis, and controlled randomness to enhance generative diversity.