This study investigates whether social class or a census-based deprivation score is a better predictor of stillbirth rates using data for 1993-5 for residents of South Thames (West) Region. Social class is routinely coded for 10% of live births and 100% of stillbirths. A Townsend deprivation score was assigned to each stillbirth and each live birth with a social class code, according to their electoral ward of residence. In unifactorial analyses of stillbirth rate the relationship was stronger with social class (P = 0.008) than with Townsend score (P = 0.11). Both relationships were strengthened by including those births recorded as social class 'other' ['other' vs. social class I odds ratio (OR) = 2.27, P < 0.001; lower vs. upper septile deprivation score OR = 1.45, P = 0.07)]. When social class and Townsend score were analysed together, the ORs for social class remained similar to before, but the Townsend ORs were reduced and non-significant overall. We conclude that social class, which is based on data on each individual, is a better predictor of stillbirth than Townsend score, which is based on data from the area of residence. We recommend further investigation of the stillbirth risk in the subgroups that make up the 'other' social class.