Abstract: To solve the problem of strength evaluation of dissimilar diffusion bonding, a strength evaluation and prediction method was proposed. Defect recognition models were built based on the binary decision tree support vector machine. The bonded ratio of the interface calculated from the recognition results was studied to establish the relationship with the shear strength. The empirical formula of shear strength was obtained by using the method of linear fitting to predict the strength of the bonding joint. It was found that the unbonded, kissing bond and pore in the dissimilar diffusion bonding interface were effectively recognized by the recognition models. The calculated bonded ratio established a good correlation with the shear strength which increased with the bonded ratio. The shear strength of dissimilar diffusion bonding can be predicted by the bonded ratio. The measured shear strength has a good agreement with the predicted shear strength.