A temperature-driven, mechanistic model predicting the development of Erysiphe necator chasmothecia in vineyards was developed and validated in 38 vineyards in the Po Valley (northern Italy), Baden-Württemberg (Germany), and South Australia between 2005 and 2011. The model, which begins operating when the first ascocarp initials are formed, predicts on a daily basis the proportions of chasmothecia at the yellow, brown and black maturity stage. The initialization date was estimated with an iterative procedure that minimized the residuals of predicted versus observed values. In all vineyards, a drop to more favourable temperatures for ascocarp production over 2–4 days in the week or in the 2 weeks before the model initialization date probably triggered chasmothecia production. Model predictions provided a good fit of observed data (coefficients of determination, model accuracy, efficacy and efficiency were all ≥0·90), with some overestimation. When predicted production of black chasmothecia (on leaves) was compared with observed dispersal of chasmothecia from vines, lack of splashing rain was probably the main cause of overestimation. When observed numbers of yellow, brown or black chasmothecia on leaves were compared with model predictions, removal of the developing chasmothecia by rainfall was probably the main cause of overestimation. Inclusion of the effect of rainfall on the removal of immature and mature chasmothecia from the powdery mildew colonies could improve the model. The model could be used to time the application of fungicides or biocontrol agents for reducing ascocarp formation and reducing primary inoculum in the following season.