Modeling and evaluation of patients' gait patterns is the basis for both gait assessment and gait rehabilitation. This paper presents a convenient and real-time gait modeling, analysis, and evaluation method based on ground reaction forces (GRFs) measured by a pair of smart insoles. Gait states are defined based on the foot-ground contact forms of both legs. From the obtained gait state sequence and the duration of each state, the human gait is modeled as a semi-Markov process (SMP). Four groups of gait features derived from the SMP gait model are used for characterizing individual gait patterns. With this model, both the normal gaits of healthy people and the abnormal gaits of patients with impaired mobility are analyzed. Abnormal evaluation indices (AEI) are further proposed for gait abnormality assessment. Gait analysis experiments are conducted on 23 subjects with different ages and health conditions. The results show that gait patterns are successfully obtained and evaluated for normal, age-related, and pathological gaits. The effectiveness of the proposed AEI for gait assessment is verified through comparison with a video-based gait abnormality rating scale.