In this work, we show the importance of multidimensional opinion representation in the political context combining domain knowledge and results from principal component analysis. We discuss the differences of feature selection between political spectrum analysis and normal opinion mining tasks. We build regression models on each opinion dimension for scoring and placing new opinion entities, e.g. personal blogs or politicians, onto the political opinion spectrum. We apply our methods on the floor statement records of the United States Senate and evaluate it against the uni-dimensional representation of political opinion space. The experimental results show the effectiveness of the proposed model in explaining the voting records of the Senate.