Flatfoot is a kind of foot deformation that can cause pain and inconvenience to people’s normal life. Early diagnosis has an important positive effect on the treatment of flatfoot. Computer-aided diagnosis (CAD) is already a relatively mature technology. We propose a new neural network, mirroring weights-and-structure-determination neural network (MWASDNN), to assist in the diagnosis of juvenile flatfoot. The experimental results show that the classification accuracy of MWASDNN for the left foot and right foot data reaches 82.35% and 84.31%, respectively. In addition, the diagnostic efficiency of the MWASDNN is very high because it takes only 0.8527 s and 0.8666 s to give the diagnostic results in left foot and right foot data.