Pregnancy in women with severe mental illness is associated with adverse outcomes for mother and infant. There are limited data on prevalence and predictors of relapse in pregnancy. A historical cohort study using anonymised comprehensive electronic health records from secondary mental health care linked with national maternity data was carried out. Women with a history of serious mental illness who were pregnant (2007–2011), and in remission at the start of pregnancy, were studied; severe relapse was defined as admission to acute care or self-harm. Predictors of relapse were analysed using random effects logistic regression to account for repeated measures in women with more than one pregnancy in the study period. In 454 pregnancies (389 women) there were 58 (24%) relapses in women with non-affective psychoses and 25 (12%) in women with affective psychotic or bipolar disorders. Independent predictors of relapse included non-affective psychosis (adjusted OR = 2.03; 95% CI = 1.16–3.54), number of recent admissions (1.37; 1.03–1.84), recent self-harm (2.24; 1.15–4.34), substance use (2.15; 1.13–4.08), smoking (2.52; 1.26–5.02) and non-white ethnicity (black ethnicity: 2.37; 1.23,4.57, mixed/other ethnicity: 2.94; 1.32,6.56). Women on no regular medication throughout first trimester were also at greater risk of relapse in pregnancy (1.99; 1.05–3.75). There was no interaction between severity of illness and medication status as relapse predictors. Therefore, women with non-affective psychosis and higher number of recent acute admissions are at significant risk of severe relapse in pregnancy. Continuation of medication in women with severe mental illness who become pregnant may be protective.