Summary: Purpose: To examine which variables available early in the course of childhood epilepsy are associated with a poor short‐term outcome and to develop models to predict such an outcome. Methods: We prospectively followed up 466 children with newly diagnosed epilepsy for 2 years. Variables were collected at intake and after 6 months. Outcome was defined as the duration of the terminal remission (TR): poor (<6 months) and not poor (≥6 months). Results: Of the subjects, 31% had a poor outcome. Multivariate analysis based on the intake variables identified number of seizures, seizure type, and etiology as risk factors for a poor outcome. With the intake and 6‐month variables combined, seizure type, etiology, the number of seizures, and not attaining a 3‐month remission during these 6 months, and the EEG at 6 months were predictive variables. A predictive model based on the multivariate logistic‐regression analysis with the intake variables was correct in 56% of the children in whom it predicted a poor outcome and in 73% of the children in whom it predicted a not‐poor outcome. With the intake and 6‐month variables together, these percentages were 66 and 79%, respectively. The sensitivity of these models was low (29 and 47%, respectively); the specificity was good (90 and 89%). Conclusions: The 2‐year outcome of childhood epilepsy is closely related to its early course. The prognosis is poor in −30% of patients. By using our data, the prediction of a poor outcome is correct in almost two thirds of the patients; however, the models produce many false‐negative predictions.