This paper presents new identifiability conditions for the Cox proportional hazard model for duration data when unobserved person specific variables are present. We compare our conditions with those presented by Elbers and Ridder. We also present identifiability conditions for a rich class of parametric hazard models without regressor variables.