In this paper we apply two statistical models to the measurement of polarization to Israeli income data over the past decade in order to empirically detect income classes as sub-populations of incomes concentrated around an optimal number of poles. The statistical models compared are a multi-resolution analysis (MRA) and a log-normal approach (LNA). We find the MRA to be superior to the LNA, by providing a more efficient allocation of households into each of the classes, reducing the overlap between the classes around the cut-values for each class. We then study polarization by use of the MRA in a multinomial logit-analysis by including ethnic-cultural, individual, family and other characteristics. We use a multiplicative normalized polarization measure developed by Palacios and Garcia (2010) which consists of presenting the interaction of three components, consistent with the axioms spelled out by Esteban and Ray (1994): alienation and identification, the number of income classes and the size distribution of the groups. The strong cultural heterogeneity of Israeli society, the sharp shifts in social policy during the observation period and the generally high quality of yearly Israeli income data render this dataset particularly useful for analyzing polarization. We find polarization to be significantly affected by cultural classes, by social policy and by standard demographic and individual characteristics. A comparison of our results with those of Esteban and Ray and Zhang and Kanbur reveals some similarity with our normalized version of Zangh and Kanbur (2001).