This paper proposes a tensor decomposition-based method for channel estimation in MIMO systems with movable antennas (MAs) at both the transmitter (Tx) and receiver (Rx). The goal is to achieve high channel estimation accuracy with low pilot training overhead. The method involves a two-stage Tx-Rx successive antenna movement pattern for pilot training, allowing the received pilot signals to be expressed as a third-order tensor. The tensor is then decomposed using canonical polyadic (CP) decomposition to estimate the parameters of multi-path channel components, including their azimuth and elevation angles and complex gain coefficients. This enables the reconstruction of the wireless channel between any pair of Tx and Rx MA positions. The uniqueness condition of the tensor decomposition is analyzed to ensure accurate and efficient channel reconstruction based on measurements at only a finite number of Tx/Rx MA positions. Simulation results demonstrate that the proposed method outperforms existing methods in terms of channel estimation accuracy and pilot overhead. The method leverages the intrinsic multidimensional characteristics of field-response-based channels, enabling efficient channel estimation with minimal pilot training overhead. The proposed approach significantly reduces pilot training overhead while maintaining high channel estimation accuracy, making it suitable for MA-enabled MIMO systems.This paper proposes a tensor decomposition-based method for channel estimation in MIMO systems with movable antennas (MAs) at both the transmitter (Tx) and receiver (Rx). The goal is to achieve high channel estimation accuracy with low pilot training overhead. The method involves a two-stage Tx-Rx successive antenna movement pattern for pilot training, allowing the received pilot signals to be expressed as a third-order tensor. The tensor is then decomposed using canonical polyadic (CP) decomposition to estimate the parameters of multi-path channel components, including their azimuth and elevation angles and complex gain coefficients. This enables the reconstruction of the wireless channel between any pair of Tx and Rx MA positions. The uniqueness condition of the tensor decomposition is analyzed to ensure accurate and efficient channel reconstruction based on measurements at only a finite number of Tx/Rx MA positions. Simulation results demonstrate that the proposed method outperforms existing methods in terms of channel estimation accuracy and pilot overhead. The method leverages the intrinsic multidimensional characteristics of field-response-based channels, enabling efficient channel estimation with minimal pilot training overhead. The proposed approach significantly reduces pilot training overhead while maintaining high channel estimation accuracy, making it suitable for MA-enabled MIMO systems.