Many of the current light-curing composite restorative techniques are rationalized in compliance with the theory that composite shrinks toward the light. Shrinkage directed toward the margins is believed to be responsible for the observed improved marginal properties. However, the dental literature does not consistently support this theory. Experimental determination of contraction patterns is very difficult. In this study, a finite element technique is used to analyze the direction of composite shrinkage as it cures. The process of polymerization can be characterized by pre- and post-gel phases. The stress developed in a restoration can be relieved quickly by the flow of material still in the pre-gel phase. Residual stresses arise after gelation. Both auto- as well as photo-curing composites were analyzed. In photo-curing composites, the gel-point varies throughout the material with the intensity of the light. Experimentally determined light transmittance data for different materials were used in the simulation. Degree of cure and time-dependent shrinkage properties were also included from experimental measurements. The analysis showed that the shrinkage direction was not significantly affected by the orientation of the incoming curing light, but instead was mostly determined by the bonding of the restoration to the tooth and by the free surfaces. Consequently, differences between the contraction patterns of auto- and photo-cure were minimal. It was concluded that composite does not shrink toward the light, but that the direction is predominantly determined by cavity shape and bond quality. Improved marginal properties should be pursued by the optimization of other factors, such as the polymerization process, the curing procedure, and the bond quality. The direction of shrinkage vectors in response to light position does not seem to be an appropriate criterion for the optimization of marginal quality.