A computational approach is presented for modeling and quantifying the structure and dynamics of the nematode C. elegans observed by time-lapse microscopy. Worm shape and conformations are expressed in a decoupled manner. Complex worm movements are expressed in terms of three primitive patterns--peristaltic progression, deformation, and translation. The model has been incorporated into algorithms for segmentation and simultaneous tracking of multiple worms in a field, some of which may be interacting in complex ways. A recursive Bayesian filter is used for tracking. Unpredictable behaviors associated with interactions are resolved by multiple-hypothesis tracking. Our algorithm can track worms of diverse sizes and conformations (coiled/uncoiled) in the presence of imaging artifacts and clutter, even when worms are overlapping with others. A two-observer performance assessment was conducted over 16 image sequences representing wild-type and uncoordinated mutants as a function of worm size, conformation, presence of clutter, and worm entanglement. Overall detected tracking failures were 1.41%, undetected tracking failures were 0.41%, and segmentation errors were 1.11% of worm length. When worms overlap, our method reduced undetected failures from 12% to 1.75%, and segmentation error from 11% to 5%. Our method provides the basis for reliable morphometric and locomotory analysis of freely behaving worm populations.