Traffic congestion is a major problem on U.S. highways. Studies and modeling of congestion can have important implications for efficiency, safety, and planning of transportation systems. A variety of theoretical models have been proposed to explain the behavior of congested traffic, including the recent interesting postulation of synchronized flow states and the treatment of these as phases (in the statistical mechanics sense). However, empirical methods for the identification and characterization of such patterns of congestion have been hampered by a lack of objective and quantitative characterizations of these patterns. An empirical study of the patterns of traffic in a data set from a congested, five-lane basic section of freeway is presented. These data were collected by videogrammetry, which provides additional information (on the spatial variation) in comparison with the information provided by other sensor technologies. With these data, a macroanalysis of the relationships among flow, density, and velocity was performed, and a linear (decreasing) relationship between velocity and density was found, with the attendant quadratic dependence of flow on density. Through a microanalysis, it was observed that depressions in velocity under congested conditions appear to be correlated with constrictions (waves of reduced speed between two waves of higher speed) that propagate backward, and a signal detection algorithm, based on convolutions of the density gradient, was designed for recognition of these events in a time series.