In this paper the definition of chaos is described at first

and then experimental rules are presented to declare whether a time series is chaotic. These rules are that time series should have an attractor with a finite dimensions

have a positive Lyapunove exponent at least

and be locally predicted. Neural network’s abiltity to restruct phase space of chaotic time series and in the condition of populared by noise is also discussed. Based on the neural network’s powerful ability of studying and nonlinear processing and local predictibility of chaos

a method to detect transient signal in the background of chaos is presented applying neural nerwork. The experimental results show this method can detect out a very weak target signal in the background of chaos.