projections of the`non-degenerate parts' of these algebras. Also, U normalizes the first algebra into the second (that is, TU Ã UT Ã Í UU Ã and T Ã UU Ã T Í U Ã U for each T P U). Conversely, we are able to characterize when U normalizes a pair of reflexive (not necessarily self-adjoint) algebras A and B. Apart from the obvious relations U Ã U Í A and UU Ã Í B, the map x must induce a bijection between the invariant projection lattices of the`non-degenerate parts' of these algebras.Thus, if U is a normalizing space then the non-degenerate parts of the von Neumann algebras generated by U Ã U and UU Ã are Morita equivalent in the sense of Rieffel [14]. Conversely, if A and B are Morita equivalent W Ã -algebras, then there are faithful representations of A and B such that the bimodule which establishes the equivalence is represented as a normalizing space U of operators between the respective Hilbert spaces. In this paper our concern is not with the notion of Morita equivalence of (abstract) W Ã -algebras, but rather with the properties of normalizers between (concrete) reflexive algebras and especially with the interplay between normalizers and reflexivity. Notice, however, that this connection between normalizers and Morita equivalence might not have been observed had we considered normalizers of a single algebra.We prove that normalizing subspaces which are bimodules over two maximal abelian self-adjoint algebras consist of operators`supported' on sets of the form f g where f and g are appropriate Borel functions. This includes the case of normalizing subspaces which are generated by operators of rank 1. In case one of the algebras U Ã U, UU Ã is abelian, the support of U turns out to be thè graph' or the`reverse graph' of a Borel function. We also show that normalizing masa-bimodules satisfy spectral synthesis in the sense of Arveson [1]. This gives a clear geometric description of the normalizers of a CSL algebra in terms of generalized graphs or partial graphs. These partial graphs are analogous to the ones appearing in the work of Feldman and Moore [7] and others. In these papers, only partial isometries normalizing certain Cartan masas are considered, while in our work the emphasis is on the whole reflexive linear space generated by each generalised graph. Also, we deal with arbitrary (non-abelian and non-self-adjoint) CSL algebras.The notation we use is standard; see, for example, [4]. We review some definitions and facts from [5] and [11]. Let H 1 and H 2 be complex Hilbert spaces, and P i the lattice of all (orthogonal) projections on H i , for i 1; 2. We let MP 1 ; P 2 denote the set of all maps J: P 1 3 P 2 which are 0-preserving and _-continuous (that is, preserve arbitrary suprema). Erdos [5] shows that each J P MP 1 ; P 2 uniquely defines semi-lattices S 1J Í P 1 and S 2J JP 1 Í P 2 such that J is a bijection between S 1J and S 2J and is uniquely determined by its restriction to S 1J . Moreover, S 1J is meet-complete and contains the identity projection, while S 2J is join-complete and contains the z...