The accurate modeling of loop segments in proteins is an important component of the overall protein folding problem. The challenge of the protein folding problem is to understand and predict the formation of the native three-dimensional structure of a protein given its primary amino acid sequence. In this paper, two methods are introduced to determine the structure of loop segments within the context of ASTRO-FOLD, an overall approach for the structure prediction of proteins. These approaches address a more difficult problem than that of traditional loop prediction in the sense that the separation distances between the loop stem regions are not assumed to be known a priori. When considering these additional degrees of freedom, the proposed methods perform extremely well, which is a result of both new modeling and algorithmic developments. In particular, the methods are validated on a testbed of benchmark protein systems, as well as a number of blind predictions from the recent CASP5 experiment. (c) 2004 Elsevier Ltd. All rights reserved.