A quantitative evaluation of neutron noise sources in a BWR plant is performed by using a multivariate autoregressive (AR) model. The mechanisms of noise propagation between the principal process signals are first examined, leading to the observations that the conventional AR model requires to be modified by means of an associate matrix in order to identify the source of flow noise. Using such a modified AR model, quantitative evaluation is made of the contribution of flow noise which is distinguished from that of void noise. Local neutron noise sources are also discussed, based on observations covering local power range monitor signals. These results indicate that flow noise, as well as void noise, contributes significantly to neutron fluctuation. The modified AR model is thus shown to be most effective for estimating the noise sources of neutron fluctuations.