The model tree algorithm of M5 is integrated with the multiple linear regression technique called modified stepwise regression (MSR), resulting in a new methodology for modeling complex civil engineering problems. We purposefully chose artificial neural networks (ANN) for comparison against the proposed “M5+MSR” because they fall at the two ends of the model interpretability spectrum in machine learning. In two application cases (case 1: concrete workability and case 2: steel fabrication estimating), the proposed “M5+MSR” gave rise to explainable regression tree models featuring substantially reduced complexities against ANN and model prediction errors comparable to ANN. In both cases, the resulting “M5+MSR” models consistently outperformed ANN in terms of model overfitting metrics by 19% in case 1 and by 21% in case 2, thus boasting better learning performances. The proposed new methodology will potentially find applications in tackling a wide range of complicated engineering problems that entail fitting prediction models based on laboratory or field data.