In many areas of engineering, the distribution of the measurements always departs from Gaussian to be heavy-tailed due to the presence of outliers, and most of the traditional identification algorithms such as the gradient based and the least-squares based algorithms are not robust in that case. This paper proposes a modified iteratively reweighted correlation analysis algorithm for robust parameter estimation of output error systems with colored heavy-tailed noises. The proposed algorithm is adopted to get the robust finite impulse response auxiliary model, and with the reconstructed noise-free output, the parameters of the output error system can be easily identified by a least squares method. The basic idea of the modified algorithm is to replace the t-distribution based m-estimator with the Tukey's biweight m-estimator, so that the outliers in a specific region can be completely rejected. Compare to the original algorithm, the modified algorithm can achieve higher estimation accuracy in Gaussian mixture noise, simulation results confirm this conclusion.