Coalmine gas concentration is a very complicated nonlinear dynamical system, which has obvious characteristics of chaos. Accurate prediction of coalmine gas concentration is very important to direct security production. However, data, got from in-wells, usually include noises, because of complex environment and multiple noises in-wells. Because of the capability of dividing frequency and diminishing noise, wavelet transformation is applied in the phase space reconstruction of chaos, which can diminish the impact of noise at the same time. In addition, the wavelet neural network is adopted to predict the concentration of coalmine gas. Experiments show that the method, based on chaotic series and wavelet neural network, has greatly improved accuracy of prediction, which is valuable for generalization and application.