This allows us, in effect, to compare the low-wage and high-wage segments after “purging” the “noisy” observations. Assigning workers to simulated segments using the maximum estimated posterior probability led to segment shares that are meaningfully different than the estimated population shares of each segment. We propose an alternative method in which observations are weighted by their posterior probabilities and then included in all simulated segments. Using this method, we found quantitatively small but qualitatively reasonable differences in the characteristics of workers between the low-wage segment and the high-wage segment. The between-segment differences in wage equation coefficients, representing the returns to these worker attributes, were much larger than the differences in worker attributes themselves. For example, Jewish females and Arab males suffer considerable wage penalties in the high-wage segment, while Arab females suffer wage penalties in the low-wage segment. Returns to schooling are considerably higher for Jewish workers in the high-wage segment, while they are positive for Arab females only in the low-wage segment, and do not exist for Arab males in either segment. Altogether, the results indicate that much of the wage disparities in Israel are due to unobserved factors rather than to observable characteristics. They also lead to some policy-relevant insights about the links between schooling, ethnic minorities and wages.