We use probabilistic, topological and combinatorial methods to establish the following deviation inequality: For any normed space X = (R n , · ) there exists an invertible linear map T :where G is the standard n-dimensional Gaussian vector and C, c > 0 are universal constants. It follows that for every ε ∈ (0, 1) and for every normed space X = (R n , · ) there exists a k-dimensional subspace of X which is (1 + ε)-Euclidean and k ≥ cε log n/ log 1 ε . This improves by a logarithmic on ε term the best previously known result due to G. Schechtman.