<p>The current paper proposes a phenomenological model to investigate martensitic variant reorientation (MVR) in ferromagnetic shape memory alloys (FSMAs). A quasi-one-dimensional lumped-element formulation is developed under the typical loading condition (i.e. uniaxial stress and a perpendicular magnetic field). MVR is delineated by a single internal state and the macroscopic coupled behaviors are determined within the framework of thermodynamics. The inherent dissipative nature of MVR is well discussed with the simplified model formulation and a feedforward artificial neural network (ANN) is employed to establish the reorientation conditions. The current ANN employs a single hidden layer with three hidden neurons. Based on the physical and geometrical considerations, the scaling policy is employed to modify the reorientation conditions when the external loads are interrupted and reversed. With the information of reversal points on the reorientation paths, the simulation of internal hysteresis loops is tackled in a memory-related manner. Besides, compressive super-elastic tests are carried out and stress-strain hysteresis loops are obtained under different bias magnetic fields. The proposed model is numerically implemented and the ANN weights and biases are identified using the measured stress vs. strain data. The predicted results are compared with the experimental counterparts and their good agreement demonstrates the model’s capabilities in addressing macroscopic coupled responses of FSMAs. The presented modeling strategy balances the computational cost and prediction accuracy, which will facilitate the design and control of FSMA-based smart structures.</p>