Urban heat waves (UHWs) are strongly associated with socioeconomic impacts. Reliable projections of these extremes are pressingly needed for local actions in the context of extreme event preparedness and mitigation. Such information, however, is not available because current multi-model projections largely lack a representation of urban areas. Here, we use a newly-developed urban climate emulator framework in combination with global climate simulations to show that, at the urban scale a large proportion of the uncertainty results from choices of model parameter and structural design in projecting UHWs in the next several decades under climate change. Omission of the model parameter and structural uncertainty would considerably underestimate the risk of UHWs. Results show that, for cities in the four high-stake regions, the Great Lakes region of North America, Southern Europe, Central India, and North China, a virtually unlikely (0.1% probability) UHW event is estimated by our model with probabilities of 13.91%, 5.49%, 2.78%, and 13.39% respectively in 2061–2070 under a high-emission scenario. Our findings highlight that for urban-scale extremes, decision-makers and stakeholders have to account for the multi-model uncertainties if decisions are informed based on climate simulations.