We use two different methods, i.e. dipole-fitting (DF) and hemisphere comparison (HC), to search for the anisotropic signals hiding in the Union2.1 data set. We find that the directions of maximum matter density derived using these two methods are about 114A degrees away from each other. We construct four Union2.1-like mock samples to test the statistical significance of these two methods. It is shown that DF method is statistically significant, while HC method is strongly biased by the distribution of data points in the sky. Then we assume that the anisotropy of distance modulus is mainly induced by the anisotropy of matter density, which is modelled to be the dipole form Omega(M) = Omega(M0)(1 - cos theta). We fit our model to Union2.1, and find that the direction of maximum matter density is well consistent with the direction derived using DF method, but it is very different from the direction previously claimed. Monte Carlo simulations show that our method is statistically more significant than HC method, although it is not as significant as DF method. The statistical significance can be further improved if the data points are homogeneously distributed in the sky. Due to the low quality of present supernovae data, however, it is still premature to claim that the Universe has any preferred direction.