We propose general quantitative methods for evaluating and visualizing the results of machine-generated style-specific accompaniment. The evaluation of automated accompaniment systems, and the degree to which they emulate a style, has been based primarily on subjective opinion. To quantify style similarity between machine-generated and original accompaniments, we propose two types of measures: one based on transformations in the neo-Riemannian chord space, and another based on the distribution of melody-chord intervals. The first set of experiments demonstrate the methods on an automatic style-specific accompaniment (ASSA) system. They test the effect of training data choice on style emulation effectiveness, and challenge the assumption that more data is better. The second set of experiments compare the output of the ASSA system with those of a rule-based system, and random chord generator. While the examples focus primarily on machine emulation of Pop/Rock accompaniment, the methods generalize to music of other genres.