A signal loss is generally reported in electron probe microanalysis (EPMA) of porous, highly divided materials like heterogeneous catalysts. The hypothesis generally proposed to explain this signal loss refers to porosity, roughness, energy losses at interfaces, or charging effects. In this work we investigate by Monte Carlo simulation all these physical effects and compare the simulated results with measurements obtained on a mesoporous alumina. A program using the PENELOPE package and taking into account these four physical phenomena has been written. Simulation results show clearly that neither porosity nor roughness, nor specific energy losses at interfaces, nor charging effects are responsible for the observed signal loss. Measurements performed with analysis of carbon and oxygen lead to a correct total of concentration. The signal loss is thus explained by a composition effect due to a carbon contamination brought by the sample preparation and to a lesser extent by a stoichiometry of the porous alumina different from a massive alumina. For this kind of high specific surface porous sample, a little surface contamination layer becomes an important volume contamination that can produce large quantification errors if the contaminant is not analyzed.