A methodology for the design of continuous practical high-gain observers for non-linear observable systems is presented. Only two conditions are required: injectivity of the observability map of order n and uniform continuity of its inverse. Using this map, the system is transformed to observability normal form, which could have a non-Lipschitz continuous right-hand side and eventually multiplicity of solutions. A so-called epsilon-approximate high gain observer is designed for this form, constituting the dynamic part of the observer. A uniformly continuous extension of the inverse of the transformation is used as the algebraic part. Convergence of the error to a neighbourhood of the origin is guaranteed.