Abstract: Leniency bias has a detrimental effect to analysis of indicators- relationships in teaching evaluation. It will distort the statistic results based on the Pear son correlation coefficient. The current paper employed polychoric correlation c oefficient to analyze teaching evaluation data, which has been proved robust and will offer more precise estimate. A non-normal real data sample shows that the re is sufficient difference between estimates of two correlation coefficient est imators. The approach based on polychoric correlation offered better fit indexes to a confirmatory factor analysis.