Summary. One problem in analysing social and demographic change through time by using census data arises from differences in the size and shape of the geographical units that are used to output data between different years. Failure to correct for changing output geographies may lead to unknown and possibly large biases when comparing results through time between different censuses. The paper addresses this issue by using the example of residential segregation in Northern Ireland. It has two main objectives. Firstly, by compiling 2001 Northern Ireland census data on 1991 census output geographies it assesses the sensitivity of indices of residential segregation to these changes in geographical units. Secondly, it suggests a method by which census analysts can assess how sensitive their results are to changing output geographies when they cannot correct for these changes and must work with the data ‘as they are’. A subsidiary aim is to contribute to the evidence base on residential segregation in Northern Ireland. The paper finds that indices of residential segregation are insensitive to changes in output geographies between 1991 and 2001. The reason suggested for this is that the units in each zonal geography are smaller than the spatial scale over which population counts are positively auto-correlated. The use of spatially weighted segregation indices is advanced as a generalizable means of learning about the geographical patterning of population in different censuses. It is argued that these insights combined with knowledge of the size of geographical units used in each census can help researchers elsewhere to judge how sensitive their results might be to changing census output geographies through time.