In the frame of bayesian inference, the inverse problem solution is described by posterior probability density function (PDF) in the model space. Estimation of PDF in a multidimensional space is based on a Markov chain having invariant probability identical to the posterior PDF of the model. Extensive exploration of the model space using a stochastic technique is expected to circumvent local minima. The Markov chain algorithm has been applied to non-linear inversion of magnetotelluric (MT) data using 1-D and 2-D models with satisfactory results, i.e. synthetic models have been recovered. Keywords: inversion, Markov chain, magnetotelluric method Hendra Grandis
