Abstract: Cotton spider mite is one of the main pests affecting cotton yield and quality. In order to quickly, accurately, and effectively monitor the occurrence of cotton spider mite, a digital image was obtained using UAV equipped with a digital camera, and a variety of visible vegetation indexes were calculated as the primary feature factors. Then, the ReliefF-Pearson feature dimensionality reduction method was used to select the best modeling features, consisting of partial least squares regression（PLSR）, BP neural network（BPNN）, and the random forest（RF） remote sensing estimation model of cotton canopy leaf chlorophyll relative content（SPAD） and remote sensing estimation model of the severity of cotton spider mite. The results showed that there was a significant negative correlation between the severity of cotton spider mites and the SPAD value of cotton canopy leaves. Through accuracy evaluation, it was determined that the RF model had the best performance, whereby the determination coefficient and root mean square error of model verification were 0.74 and 2.13, respectively. The results showed that the remote sensing estimation model of SPAD value of cotton canopy leaves can accurately estimate the damage of cotton spider mites, and provide a reference basis for non-destructive monitoring and pest control of cotton spider mites.