Classical peak detection methods, used for event detection in intensified CCDs, do not permit the location of events in two adjacent pixels of a single frame, dramatically reducing the spatial resolution of the observations. To overcome this limitation, in the context of real time processing, a Hopfield neural network for event detection is proposed. A n2 neuron network is designed so that each neuron is associated with a corresponding pixel of a n X n window scanning the CCD frames. The n2 pixel values of the window are inputs to the network which evolves dynamically according to the propagation rules of the Hopfield model. At convergence, the activation state of each neuron is a binary value, 0 or 1, operating as an event flag for the associated pixel. The network parameters are determined by formulating the event detection problem as a signal decision problem, assuming a model of shape for the photon splashes in the detector. Digital implementation of the network has been studied and simulation results are presented.