Objective To dissect specific gait abnormalities associated with upper motor neuron (UMN) dysfunction in amyotrophic lateral sclerosis (ALS) by controlling for overall disease severity and to develop a multivariate classification model. Methods We performed 3D gait analysis on 118 ALS patients and 1796 healthy controls (HC). ALS patients were categorized into those with ALS with UMN dysfunction((ALS-UMN), n = 70) and those without ALS without UMN signs ((ALS-Numn), n = 48) lower limb UMN signs based on neurological examination. Gait parameters were compared, and their association with UMN involvement was analyzed using partial correlation (controlling for ALSFRS-R score) and machine learning models (Random Forest and Least Absolute Shrinkage and Selection Operator (Lasso) regression). Results Compared with HC, ALS patients exhibited widespread gait deterioration (e.g., reduced speed, increased step width, p < 0.001). After controlling for ALSFRS-R, specific parameters, including reduced stride, increased step width, prolonged double support, and elevated gait cycle time asymmetry, remained independently associated with UMN severity (PENN score, p < 0.01). A multivariate model incorporating key features demonstrated fair discriminative ability for identifying ALS-UMN patients, with an area under the curve (AUC) of 0.690, a sensitivity of 0.816, and a specificity of 0.418. Conclusion Quantitative gait analysis reveals a distinct spatiotemporal pattern linked to UMN dysfunction in ALS. A model based on gait features shows potential, particularly high sensitivity, for identifying patients with pyramidal signs, supporting the exploratory utility of objective gait metrics for motor phenotyping in ALS, pending external validation.