In this Letter, a novel hybrid representation of protein structure is proposed by utilizing two sources of information. One is the distribution of C-alpha-C-alpha distances with sequence separation three, which describes the local geometry property and is used to identify contents of regular secondary structures; the other is the linear sequence distance distribution of medium and long range interactions, which represents packing arrangement and topological connections between secondary structures. Furthermore, we introduce a new protein structure comparison method based on information theory. Cluster analysis and structure classification experiments on several data sets demonstrate its effectiveness on measuring protein fold similarity. (c) 2006 Elsevier B.V. All rights reserved.