A radial basis function artificial neural network (RBF ANN) based method for distributed dynamic force reconstruction is presented in this study considering multi-source uncertainties. Foremost, the distributed forces are approximated by truncated Legendre orthogonal polynomial in the time history. The RBF ANN implementation is introduced, whereby the normalized structural responses are deemed as input variables, and the normalized polynomial coefficients are regarded as output variables. To improve the generalization of the trained RBF ANN, the structural responses and polynomial coefficients with respect to different types of distributed dynamic forces at every instant are defined as the samples. Thus, the coefficients of the distributed force to be identified may be ascertained after inputting the corresponding responses. In terms of the multi-source uncertainties, the random white Gaussian noise is brought in to simulate the noise of measured signals polluted, and the interval vector is involved to characterize the uncertainties of material dispersion subsequently. In view of the issue of large interval parameters, the subinterval theory is established by a combination of the Taylor series expansion in each subinterval. Eventually, the validity and feasibility of the developed methodology are demonstrated by two numerical examples.