The prevalent Mathematical Programming Neural Network (MPNN) models are surveyed, and MPNN models have been developed and applied to the unconstrained optimization of mechanisms. Algorithms which require Hessian inversion and those which build up a variable approach matrix, are investigated. Based upon a comprehensive investigation of the Augmented Lagrange Multiplier (ALM) method, new algorithms have been developed from the combination of ideas from MPNN and ALM methods and applied to the constrained optimization of mechanisms. A relationship between the weighted least square minimization of design equation error residuals and the mini-max norm of the structure error for function generating mechanisms is developed and employed in the optimization process; as a result, the computational difficulties arising from the direct usage of the complex structural error function have been avoided. The paper presents relevant theory as well as some numerical experience for four MPNN algorithms.