Abstract. It is known that the basic tensor valuations which, by a result of S. Alesker, span the vector space of tensor-valued, continuous, isometry covariant valuations on convex bodies, are not linearly independent. P. McMullen has discovered linear dependences between these basic valuations and has implicitly raised the question as to whether these are essentially the only ones. The present paper provides a positive answer to this question. The dimension of the vector space of continuous, isometry covariant tensor valuations, of a fixed rank and of a given degree of homogeneity, is explicitly determined. The approach is constructive and permits one to provide a specific basis. §1. Introduction Alesker [6,7] determined these spaces and their dimensions explicitly.As a natural generalization of the motion invariant real valuations, McMullen [14] introduced isometry covariant tensor valuations, and he formulated the aim to find a characterization of such valuations, under continuity assumptions. To explain this, we denote by T r the vector space of symmetric tensors of rank r ∈ N 0 (the nonnegative integers) over R n (we use the scalar product of R n to identify R n with its dual space). The symmetric tensor product of symmetric tensors a, b is denoted by ab. We write x r for the r-fold symmetric tensor product of x ∈ R n . The normalization is chosen so that x r = x ⊗ · · · ⊗ x (with r factors x). A tensor valuation on K n is a valuation on K n 2000 Mathematics Subject Classification. Primary 52A20.