This paper focuses on the estimation of mutual fund styles by return-based style analysis. Usually, the investment style is assumed to be either constant through time, or time variation is implicitly accounted for by using rolling regressions. The former assumption is often contradicted by data analysis, and the latter is inefficient due to its ad hoc chosen window size. We propose to use the Kalman filter to explicitly model time-varying exposures of mutual funds. This leads to a testable model and more efficient use of the data, which reduces the influence of spurious correlation between mutual fund returns and style indices.