<p>Magnetotelluric (MT) data inversion is a complex problem with highly nonlinear characteristics. The one dimensional (1-D) MT inversion algorithm adopts the linear or pseudo-linear method, which can only reflect the approximate model morphology, resulting in its inability to meet the requirements of high-precision inversion. To solve this problem, a cuckoo search (CS) algorithm is proposed for MT inversion, in which the optimization objective function is jointly composed of apparent resistivity and impedance phase data. First, a 1-D stratified geoelectric model is established, MT inversions for the four-layer (HK-type) and five-layer (HKH-type) geoelectric model is performed using the CS technique, particle swarm optimization (PSO) algorithms and the damped least squares (DLS) imaging method, and the effect of random noise (5% and 10% Gaussian noise) on the inversion accuracy is analyzed. The numerical experimental results show that the proposed approach has a high degree of model fitting, great anti-noise performance and fast convergence. Then, two type pseudo 2-D model inversions using the CS algorithm are established, and the results prove that the proposed method can accurately reflect the distribution of anomalous bodies. Finally, the effectiveness of CS for MT data inversion was verified by field data and compared with traditional PSO algorithm. The research results provide new methods and techniques for MT data interpretation.</p>