Aerosol is an important component of the Earth's atmosphere, affecting weather, climate and diverse elements of the biosphere. Satellite sounders are an essential tool for measuring the highly variable distributions of atmospheric aerosol. Here, we present a new algorithm for estimating atmospheric dust optical depths and associated retrieval uncertainties from spectral radiance measurements of the thermal infrared sounder IASI. The retrieval is based on the calculation of a dust index and on a neural network trained with synthetic IASI spectra. It has an inherent high sensitivity to dust and efficiently discriminates dust from other aerosols. In particular, over remote dust‐free areas, the retrieved levels of optical depth have a low bias. Over sea, noise levels are markedly lower than over land. Performance over deserts is comparable to that of other land surfaces. We use ground‐based coarse mode aerosol measurements from the AERONET network to validate the new product. The overall assessment is favorable, with standard deviations in line with estimated uncertainties, low biases and high correlation coefficients. However, a systematic relative bias occurs between sites dominated by African and Asian dust sources respectively, likely linked to differences in mineralogy. The retrieval has been performed on over a decade of IASI data, and the resulting dataset is now publicly available. We present a global seasonal dust climatology based on this record, and compare with those obtained from independent satellite measurements (MODIS and a third‐party IASI‐product) and dust optical depth from the ECMWF model.