This article questions the slope homogeneity in a gravity equation and proposes a partially heterogeneous framework for its estimation using panel data. We suggest to employ K-mean clustering to group countries according to the gravity equation variables. Further, the gravity model is estimated on these created homogeneous groups. We apply this procedure on analysis of German trade data and confirm the slope heterogeneity in the model. When we estimate the model on each cluster separately, the estimated coefficients and their standard errors vary sufficiently. Moreover, we show that the pooled estimation technique severely under- or overestimate the effect of given variables.