The wakes behind a square cylinder were simulated using two-equation turbulence models, k-ε and RNG k-ε models. For comparisons between the model predictions and analytical solutions, we employed three skill assessments:, the correlation coefficient for the similarity of the wake shape, the error of maximum velocity difference (EMVD) of the accuracy of wake velocity, and the ratio of drag coefficient (RDC) for the flow patterns as in the authors’ previous study. On the basis of the calculated results, we discussed the feasibility of each model for wake simulation and suggested a suitable value for an eddy viscosity related constant in each turbulence model. The k-ε model underestimated the drag coefficient by over 40 %, and its performance was worse than that in the previous study with one-equation and mixing length models, resulting from the empirical constants in the ε-equation. In the RNG k-ε model experiments, when an eddy viscosity related constant was six times higher than the suggested value, the model results were yielded good predictions compared with the analytical solutions. Then, the values of EMVD and RDC were 3.8 % and 3.2 %, respectively. The results of the turbulence model simulations indicated that the RNG k-ε model results successfully represented wakes behind the square cylinder, and the mean error for all skill assessments was less than 4 %.