Let G be a finite group and V, V finite dimensional real orthogonal representations of G. Then V is said to be topologically similar to V (V ∼ t V ) if there exists a homeomorphism h : V → V which is G-equivariant. If V, V are topologically similar, but not linearly isomorphic, then such a homeomorphism is called a nonlinear similarity.The topological classification of G-representations was first studied by de Rham [18]. He proved that if a topological similarity h : V → V of orthogonal representations preserves the unit spheres and restricts to a diffeomorphism between S(V ) and S(V ), then V and V are linearly isomorphic. The purpose of this paper is to give new restrictions on the existence of non-linear similarities using techniques from bounded topology.Theorem A. Let V, V be real orthogonal G-representations, where G is a finite cyclic group. Suppose that Res H V ∼ = Res H V for each proper subgroup H G, and thatThis result gives information about topological similarities for non-cyclic groups as well, since linear equivalence of representations is detected by character values. The formulation is also well-adapted to inductive arguments, and we get a new proof for the results of [10], [12].Corollary B. Let G be a finite group and V, V be real orthogonal G-representations. Suppose that for each cyclic subgroup C ⊆ G of 2-power order, the elements of C act All previously contructed topological similarities of cyclic groups G contain the non-trivial 1-dimensional representation (i.e. the representation with isotropy group H ⊂ G of index 2), or are induced from such examples.For example, suppose that G = Z/4q, q = 2 r−2 , r ≥ 4. Now let V 1 = t i + t j , with i ≡ 1 mod 4 and j ≡ ±i mod 8, and V 2 = t i+2q + t j+2q where t denotes a faithful 2-dimensional representation of G. Let (resp. δ) denote the 1-dimensional trivial (resp. non-trivial) representation of G. ThenG, so our result says that V 1 ⊕ W is not topologically similar to V 2 ⊕ W unless δ is a summand of W . More generally,is also a non-linear similarity and the non-trivial 1-dimensional H-representation is a summand of Res H W .Our techniques also give information about the existence and classification of nonlinear similarities. We recently noticed that [4, Thm 1(i)] is incorrect as stated. For example for G = Z/12, there are no 6-dimensional non-linear similarities. The problem is that the natural epimorphism π : Z/4q → Z/2 r where q = 2 r−2 s, s odd, does not induce an isomorphism π * :as claimed in [4, p. 732 l. -8]. These topics will be discussed elsewhere [8].