A new experimental/numerical technique of classification of flow regimes (flow patterns) in air-magnetic fluid two-phase flow is proposed in the present paper. The proposed technique utilizes the electromagnetic induction to obtain time-series signals of the electromotive force, allowing us to make a non-contact measurement. Firstly, an experiment is carried out to obtain the time-series signals in a vertical upward air-magnetic fluid two-phase flow. The signals obtained are first treated using two kinds of wavelet transforms. The data sets treated are then used as input vectors for an artificial neural network (ANN) with supervised training. In the present study, flow regimes are classified into bubbly, slug, churn and annular flows, which are generally the main flow regimes. To validate the flow regimes, a visualization experiment is also performed with a glycerin solution that has roughly the same physical properties, i.e., kinetic viscosity and surface tension, as a magnetic fluid used in the present study. The flow regimes from the visualization are used as targets in an ANN and also used in the estimation of the accuracy of the present method. As a result, ANNs using radial basis functions are shown to be the most appropriate for the present classification of flow regimes, leading to small classification errors.