Existing tests to measure the ability to recognize other people's emotional expressions (emotion recognition ability [ERA]) mostly focus on a single modality (usually the face) and include only a small number of emotions, restricting their ecological validity. Further, their reliability is often unsatisfactory. The goal of the present study was to develop a new ERA test (Geneva Emotion Recognition Test [GERT]) that (a) features dynamic and multimodal actor portrayals (short videos with sound), (b) contains a large number of emotions, and (c) is based on modern psychometric principles (item response theory). We asked 295 participants to watch 108 actor portrayals and to choose, for each portrayal, which of 14 emotions had been expressed by the actor. We then applied the Rasch model independently to each of the 14 emotion portrayal subsets to select 83 final items for the GERT. Results showed that the model fits the emotion subtests and the overall GERT and that measurement precision is satisfactory. Consistent with previous findings, we found a decline in ERA with increasing age and an ERA advantage for women. To conclude, the GERT is a promising instrument to measure ERA in a more ecologically valid and comprehensive fashion than previous tests.