Abstract: A dim small moving target detection algorithm based on joint spatio-temporal sparse recovery is proposed in this paper. A spatio-temporal over-complete dictionary is firstly trained from infrared image sequence, and it can characterize not only motion information but also morphological feature. In the spatio-temporal over-complete dictionary, the spatio-temporal atom are then classified as target spatio-temporal atoms building target spatio-temporal over-complete dictionary, which describes moving target, and background spatio-temporal atoms constructing background spatio-temporal over-complete dictionary, which embeds background clutter. Infrared image sequence is decomposed on the union of target spatio-temporal over-complete dictionary and background spatio-temporal over-complete dictionary. The residuals after decomposing and reconstruction by the target spatio-temporal over-complete dictionary and background over-complete dictionary differ very distinctly, and they are then adopted to decide the signal is from target or background. Some experiments are conducted and the experimental results show that the residual reconstructed by its homologous spatio-temporal over-complete dictionary is very little, yet the residual recovered by its heterogonous spatio-temporal over-complete dictionary is quite large. This proposed approach could not only improve the sparsity more efficiently, but also enhance the target detection performance more effectively.