Satellite images of land surface temperatures (LST) are commonly used to identify areas within cities most prone to diurnal thermal discomfort, but they may not reflect the experiences of pe- destrians. Here, we developed predictive statistical models for Physiological Equivalent Tem- perature (PET), an indicator of thermal discomfort, with easily accessible spatial predictors. For this, we measured PET (n = 4472) along eight transects (range: 700–5000 m) using a multi-sensor instrument in the urban fabric of Geneva, Switzerland during periods of summer heat. We par- ametrised generalised additive models (GAM) and linear mixed models (LMM) with six commonly available predictor variables [solar energy, Local Climate Zone (LCZ), albedo, LST, Normalized Difference Moisture Index (NDMI) and canopy cover]. We found that LST, alone, explained <2 % of observed variation in PET, whereas the GAM with all the 6 predictor variables had R2 = 0.43. LCZ and solar energy explained most of the variability of PET across the city. PET values were lower in the densely built city centre than in the peri-urban environment. LST is poorly correlated with air temperature and PET in urban settings, and thus should not be used alone to predict outdoor thermal discomfort.