Protein fold recognition has been approached by threading an amino acid sequence onto a library of folds, calculating a sequence-structure compatibility score, and ranking these scores. Due to imperfections in the empirically derived pairwise potential functions and the necessarily heuristic approach to the sequence-structure alignment problem, the method benefits from the assessment of threaded models to evaluate the most probable structures among the list of possible folds. THREADER and ANALYST, software tools available through the Internet, facilitate the alignment and assessment steps of a threading prediction. No process has been found to be universally reliable for the detection of folds related to the structure of a known input sequence, but several useful steps and approaches are discussed.