In this article, we present an observational investigation of ground motion at Mexico City focused on surface waves. Our purpose is 2-fold; first, to understand incident ground motion during the great Michoacán earthquake of 19 September 1985, and second, to characterize surface waves propagating in the lake-bed zone. To this end we analyze the strong-motion records obtained at Mexico City for the large (MS = 8.1) earthquake of 19 September 1985. It is shown that, in the low-frequency range, we observe the Rayleigh fundamental mode in both the vertical and the radial components, and the Love fundamental mode in the transverse component at all the strong-motion stations. The vertical component also shows the first higher mode of Rayleigh waves. We use a very broadband record obtained at station CU for the smaller (MS = 6.7) earthquake of 14 May 1993 to verify that the dispersion computed from the model of Campillo et al. (1989) represents well the average surface-wave propagation between the coast and Mexico City in the 7- to 10-sec period range. We use this result to assign absolute times to the strong-motion records of the Michoacán event. This allowed us to identify additional wave trains that propagate laterally in directions other than great circle in the 3- to 5-sec period range. These wave trains are identified as Love waves. In a second analysis, we study a set of refraction data obtained during a small-scale (250 m) experiment on the virgin clay of the lake-bed zone. Phase-velocity dispersion curves for several modes of Rayleigh waves are identified in the refraction data and inverted to obtain an S-wave velocity profile. This profile is used as the uppermost layering in a 2D model of Mexico City valley. The results of numerical simulation show that surface waves generated by lateral finiteness of the clay layer suffer large dispersion and attenuation. We conclude that surface waves generated by the lateral heterogeneity of the upper-most stratigraphy very significantly affect ground motion near the edge of the valley, but their importance is negligible for distances larger than 1.5 km from the edge. Thus, locally generated surface waves propagating through the clay layer cannot explain late arrivals observed for the 1985 event. We suggest that the long duration of strong motion is due to the interaction between lateral propagation of waves guided by deep layers (1 to 4 km) and the surficial clay layer. This interaction is possible by the coincidence of the dominant frequency of the uppermost layers and the frequency of the deeply guided waves.