Stress at work and its consequences are of growing interest. The Munich Employee Health Questionnaire (MEHQ; Zweck, 2017) was developed to measure both psychosocial hazards and outcomes considering health (strains, general health, sickness absence) and turnover intention. After reviewing psychosocial hazards that are supposed to be associated with stress, we show estimates of criterion validity for the MEHQ from cross-sectional and longitudinal (more than one year) analyses, based on online panel data from n = 1,327 German employees (sample sizes for longitudinal analyses between n = 383 and n = 444). Using methods from machine learning and predictive modeling in addition to linear regression, we could predict all outcomes except sickness absence both cross-sectionally and longitudinally. Highest predictive performance (estimated by 10 times repeated 10-fold cross-validation) was achieved for the outcome strains (cross-sectional: R2 approximately .5, longitudinal: R2 approximately .4). Longitudinal predictions could be greatly improved when outcomes measured at time 1 (among other covariates) were added as predictors. General health, strains, and turnover intention could be predicted using either the MEHQ domain sum scores or items as predictors directly. Classical linear regression models using sum scores showed predictive performance comparable to elastic net models with all items as predictors. No performance improvement was observed when using the nonlinear random forest. The MEHQ seems to be a competitive measure for predicting strains, general health, and turnover intention. Reasons for the low predictability of sickness absence are discussed.