Apparel manufacturing is an industry with high energy consumption and carbon emissions. With the development of the low-carbon economy, low-carbon production in the apparel manufacturing industry become more and more imperative. The apparel industry is encountering great challenges in reducing carbon emissions. Garment sewing comprises a large number of processes, machines and operators. However, the existing studies lack quantitative analysis of carbon sources in the sewing process. This study analyzed the carbon emission characteristics in garment sewing production. Evaluation models of carbon emission were established for the sewing process in this research and the factors of fabrics, accessories, sewing machines and operators were included in the models. The results showed that fabrics and accessories were the main sources of carbon emissions in garment sewing production. The second largest carbon emission source was sewing machines, followed by operators. According to the evaluation models, the number of machines, operators and the utilization rate of the machines were related to the balance of the assembly line. A multi-objective optimization model aimed at minimizing the time loss rate and smoothness index of the assembly line was established, and a fast and elitist multi-objective genetic algorithm was used to obtain the solution for carbon emission reduction. The men’s shirt assembly lines, based on three types of workstation layouts (the order of processes, the type of machines and the components of the garment), were applied to verify the effectiveness of the model and algorithm. The results indicated that the total carbon emissions of the three assembly lines based on balance optimization were less than that of the normal assembly line. The assembly line of the workstations arranged in the order of processes was the best assembly line since it had the highest efficiency and the lowest carbon emissions.