Soil water retention curve reflects the basic relationship between the mass and energy of water in vadose zone, and indirectly reflects soil pore size distribution. The fitting accuracy of soil water retention curve model depends not only on the model, but also on the selection of the fitting algorithm. In order to obtain the most suitable combination of model and fitting algorithm, a variety of fitting algorithms were used to investigate the fitting performance based on a soil database (UNSODA 2.0). Four widely used soil water retention models (i.e., Brooks-Corey model, van Genuchten model, Kosugi model and Biexponential model) and four fitting algorithms (i.e., genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm and differential evolution algorithm) were selected to compare the differences among the models. Results show that van Geunchten model has excellent performance in fine and coarse textured soils. Biexponential model outperforms in medium textured soils, while Brooks-Corey model has poor overall fitting effect. Particle swarm optimization algorithm and differential evolution algorithm have the best fitting performance in fitting soil water retention curves. The computing efficiency of simulated annealing algorithm is more efficient than the other algorithms. When the amount of data to be processed is large, this algorithm can significantly reduce the operation time. The results of this study can provide some reference for the selection of soil hydraulic parameters in the field of farmland irrigation and ecological hydrology.