We propose maximum likelihood (ML) and restricted maximum likelihood (REML) methods for estimating the mean and covariance parameters of multi-input multi-output (MIMO) Ricean and Rayleigh block-fading channels using measurements from multiple coherent intervals containing both amplitudes and phases of the received signal. Correlated and independent fading scenarios with structured and unstructured line-of-sight (LOS) array response models are considered. Computationally efficient ML and approximate ML (AML) estimators are proposed for unitary space-time modulation schemes and orthogonal designs in correlated fading. We also derive Crame/spl acute/r-Rao bounds (CRBs) for the unknown parameters, discuss initialization of the proposed algorithms, and evaluate their performance via numerical simulations under the block- and continuous-fading scenarios.