Abstract. Pairs of 2-block Toeplitz (N × N )-matrices (T s ) ij = p 2i−j+s−1 , s = 0, 1, i, j ∈ {1, . . . , N}, are considered for arbitrary sequences of complex coefficients p 0 , . . . , p N . A complete spectral resolution of the matrices T 0 , T 1 in the system of their common invariant subspaces is obtained. A criterion of nondegeneracy and of irreducibility of these matrices is derived, and their kernels, root subspaces, and all common invariant subspaces are found explicitly. The results are applied to the study of refinement functional equations and also subdivision and cascade approximation algorithms. In particular, the well-known formula for the exponent of regularity of a refinable function is simplified. A factorization theorem that represents solutions of refinement equations by certain convolutions is obtained, along with a characterization of the manifold of smooth refinable functions. The problem of continuity of solutions of the refinement equations with respect to their coefficients is solved. A criterion of convergence of the corresponding cascade algorithms is obtained, and the rate of convergence is computed. §1. IntroductionThe finite-dimensional 2-Toeplitz operators corresponding to a given sequence of complex numbers p 0 , . . . , p N are the operators T 0 , T 1 acting in R N and determined by the following matrices:(we set p k = 0 for k < 0 and for k > N). Usually, the matrices T 0 , T 1 are referred to in the literature as 2-block, or (2 × 1)-block, or two-slanted Toeplitz matrices. For simplicity, in the sequel we call them simply Toeplitz matrices, although this differs from the conventional definition. The linear operators corresponding to the matrices (1) will be denoted by the same symbols and will also be called Toeplitz operators. Without loss of generality, at the expense of a possible shift of the argument and a change of the dimension, it may be assumed that p 0 p N = 0. For example, if N = 6, then the Toeplitz matrices are