Wood failure percentage (WFP) is an important index for evaluating the bond strength of plywood. Currently, the method used for detecting WFP is visual inspection, which lacks efficiency. In order to improve it, image processing methods are applied to wood failure detection. The present study used thresholding and K-means clustering algorithms in wood failure detection, and four kinds of plywood were manufactured to analyze the influences of wood species and adhesive. Results show that the detection by K-means clustering method is more accurate than thresholding method; it could better correlate with visual inspection results, while the detection results by thresholding method could not reflect the fluctuation of visual inspection results with types of plywood. Moreover, both analyses of the influence of adhesive and wood species show that thresholding method based detection results are more affected by adhesive color, veneer color and permeability of poplar and eucalyptus veneer (mean absolute error compared with visual inspection: PF-Eucalyptus: 15.77 %; PF-Poplar: 25.42 %; UF-Eucalyptus: 30.55 %; UF-Poplar: 21.48 %); whereas K-means clustering method based detection results show no significant change as adhesive and wood species varies (PF-Eucalyptus: 11.07 %; PF-Poplar: 10.22 %; UF-Eucalyptus: 14.77 %; UF-Poplar: 8.50 %). It can be concluded that K-means clustering method has better compatibility for different adhesive and wood species in wood failure detection.