Evidence is unclear as to whether there is a socio-economic gradient in cerebral palsy (CP) prevalence beyond what would be expected from the socio-economic gradient for low birthweight, a strong risk factor for CP. We conducted a population-based study in five regions of the UK with CP registers, to investigate the relationship between CP prevalence and socio-economic deprivation, and how it varies by region, by birthweight and by severity and type of CP. The total study population was 1 657 569 livebirths, born between 1984 and 1997. Wards of residence were classified into five quintiles according to a census-based deprivation index, from Q1 (least deprived) to Q5 (most deprived). Socio-economic gradients were modelled by Poisson regression, and region-specific estimates combined by meta-analysis. The prevalence of postneonatally acquired CP was 0.14 per 1000 livebirths overall. The mean deprivation gradient, expressed as the relative risk in the most deprived vs. the least deprived quintile, was 1.86 (95% confidence interval [95% CI 1.19, 2.88]). The prevalence of non-acquired CP was 2.22 per 1000 livebirths. For non-acquired CP the gradient was 1.16 [95% CI 1.00, 1.35]. Evidence for a socio-economic gradient was strongest for spastic bilateral cases (1.32 [95% CI 1.09, 1.59]) and cases with severe intellectual impairment (1.59 [95% CI 1.06, 2.39]). There was evidence for differences in gradient between regions. The gradient of risk of CP among normal birthweight births was not statistically significant overall (1.21 [95% CI 0.95, 1.54]), but was significant in two regions. There was non-significant evidence of a reduction in gradients over time. The reduction of the higher rates of postneonatally acquired CP in the more socioeconomically deprived areas is a clear goal for prevention. While we found evidence for a socio-economic gradient for non-acquired CP of antenatal or perinatal origin, the picture was not consistent across regions, and there was some evidence of a decline in inequalities over time. The steeper gradients in some regions for normal birthweight cases and cases with severe intellectual impairment require further investigation.