Most studies about the geographical location of a phenomenon or a population first aggregate data according to administrative boundaries which are generally not related to the issue. When the population under study is rare, and when the sample is not exhaustive, these aggregation choices often lead the results. In this study, we use non-parametric kernel techniques to estimate ratios of population densities. Using this kind of method is not technically costly and allows one to obtain the optimal trade-off between variance and bias. We apply this technique to the distribution of immigrants over the French territory, using data form population censures between 1968 to 1999. Another contribution of this work is to propose a concentration index based on the density ratios obtained in the first step.