Abstract: Due to the influence of infrared focal plane arrays production technology and material characteristics, blind pixels are inevitable in infrared focal plane arrays, which seriously affects the processing and application of infrared data. The push-broom thermal infrared hyperspectral imager which using grating system generally takes one dimension of the infrared focal plane arrays as the spectral dimension and the other dimension as the spatial dimension, which is quite different from the imaging mechanism of the thermal imager with two spatial dimensions. Conventional laboratory calibration and scene detection methods based on moving window cannot meet the requirements of blind pixel detection on the thermal infrared hyperspectral imager. A new blind pixel detection algorithm based on spectral angle matching was proposed to detect the blind pixels in thermal infrared hyperspectral imager. Taken spectral dimension information into account, this method generated temperature rise spectrum data from blackbody calibration data at different temperatures. Based on the basis of data regularization processing, the pseudo-spectral curve of effective pixels were extracted automatically, and the blind pixels were detected automatically by means of spectral angle matching. To validated the new blind pixel detection algorithm, a typical thermal infrared hyperspectral imager was used to collect image data and detecte the blind pixels of the imager. The results indicates that this method makes full use of spectral information of the thermal infrared hyperspectral imager and has high detection accuracy. The data after blind pixels compensation can satisfy the application of thermal infrared hyperspectral data.