The New Zealand Longitudinal Census (NZLC) cohorts link data for individuals across six New Zealand Censuses from 1981–2006. Approximately 70-76% of eligible individuals were able to be linked from each Census to the previous one, while 31% of eligible individuals (n~647,000) were able to be linked across all six Censuses. While the creation of longitudinal cohorts from Census data provides some wonderful research opportunities, the incomplete linkage means there is the potential for bias if associations among those linked differ from associations in the full population. I will describe the NZLC cohorts and also our attempts to identify and adjust for bias. I will conclude that while there are steps that can be taken to reduce bias, it is difficult to remove it entirely. However, unaccounted for bias tends to be small, especially in analyses with many covariates.