In this paper we present a new method for linear and nonlinear sensor network design with applications to variable estimation based on the use of accuracy and residual accuracy. Accuracy has been shown recently to be a better substitute for existing data-reconciliation network properties, such as precision, residual precision, gross error detectability and resilience as pointed out in our previous works (the first two references). We explore the numerical difficulties associated to different ways of calculating accuracy and different methods used to optimize the cost.