Studies on voter ideology usually rely on surveys where respondents place themselves on a left-right scale. This approach has three apparent problems: First, respondents may not understand the meaning of “left” and “right”. Second, they may have a biased view of their own position. Third, a unidimensional axis may not suffice to describe a given ideology coherently. I tackle all three problems by investigating how experts would perceive the ideology of each voter by applying machine learning and classical regression analysis to data from the Chapel Hill Expert Survey and the European Voter Study. Random forest regression outperforms other approaches in terms of i) in-sample fit, ii) out-of-sample forecast and iii) predicted ideological difference to the party voted for in the last national elections. My analysis suggests that there is a significant and sizeable “center bias”, i.e. voters are much more likely to place themselves at the political center than experts are predicted to do. Nevertheless, the predicted level of ideological fragmentation is lower than the fragmentation based on self-reported ideology. Departing from a unidimensional ideological axis, I show that voters tend to be more left-wing economically than generally, and that the ideological fragmentation along the economic axis is lower than along an “authoritarian-libertarian” axis. This paper shows that machine learning can be a fruitful tool to predict the political landscape and points to directions for future research.