This study conducts a data-driven classification of the response styles for the 2,131 respondents of the SHARE (Survey of Health, Ageing and Retirement in Europe) survey. In the standard Likert scale measurement, item responses reflect not only the latent traits of respondents but also their response style biases which are irrelevant for the purpose of the original measurement. The anchoring vignettes is an effective method to measure and correct for such biases. In this study, we first modeled the anchoring vignettes variables in the SHARE dataset using the Bayesian multidimensional item response model. Then, we classified the estimated individual response style parameters using the divisive analysis clustering. As a result, seven different clusters of response styles were obtained. While some of them correspond to the well-documented response styles, many of the clusters of respondents exhibit unique response styles which are both interpretable and relevant. Thus, bottom-up classification approach of response styles would undertake a key role in revealing the empirical analysis of item response behavior.