In Epidemiology, no simple test is available for comparing an observed number of exposed person-years with an expected number. The number of person-years is a sum of lengths of stays which is not normally distributed under the null hypothesis. A method is proposed and an example is given. The objective was to study in the French territory of New-Caledonia the relationship between past places of living and occurrence of respiratory cancers. The total observed number of person-years in a zone, among cases, was compared to an expected distribution, based on censuses and controlling for two confounding factors, period of stay and ethnic group. Different methods are compared: a test based on a simulated distribution, and parametric approaches with normality assumption. If the number of cases is small, the simulation test seems to be the best approach.