Analyzing the long-term evolution of meso-scale urban morphological types is constrained by the “small-sample, high-dimensional” nature of historical data, which weakens the robustness and interpretability of conventional data-driven methods and limits the use of morphological evidence in digital urban analysis and built environment governance. To address this, we propose a theory-guided modular principal component analysis (TG-MPCA) framework for sample-constrained morphological research. By embedding domain knowledge into dimensionality reduction, the framework extracts compact, low-dimensional morphological indices that are both morphologically interpretable and internally stable within the selected sample set. Applied to Harbin’s representative districts through unsupervised hierarchical clustering, evolutionary lineage analysis, and transition node identification, it reveals a “layered response pattern” among morphological modules, an asymmetrical mapping between morphological types and historical periods, a structural breakpoint around 1946 that parallels post-war Western urban restructuring, and anomalous deceleration and premature convergence in typological evolution during transitional periods. These observations offer a meso-scale morphological perspective for understanding both the spatial transformation of Chinese cities and the developmental challenges of Northeast China’s old industrial bases, while demonstrating the value of theory-guided quantitative analysis for transforming fragmented historical spatial information into interpretable morphological evidence in data-constrained contexts.