The delamination detection problem is formulated as an optimization problem with mixed type design variables using genetic algorithms. Natural frequency is taken as the global damage index for detection of delamination in composite laminates. A recently developed finite element model based on an improved layerwise composite laminate theory is employed to calculate the natural frequencies of laminates with given delamination pattern. Artificial backpropagation neural networks are trained to simulate results from the finite element analysis. These artificial neural networks are chosen as function approximations which are used to predict natural frequencies of delaminated laminates with satisfactory accuracy. Two different types of delamination configurations - through-the-width delamination and internal delamination, have been considered in the present studies. A new modular approach has been developed in constructing the backpropagation neural networks for the internal delamination problem. In this approach, instead of a single backpropagation neural network, multiple smaller size module backpropagation neural networks are used to simulate the vibration modal response. Results with satisfactory accuracy have been obtained in detecting the internal delamination using this technique.