Wildfire spread in living vegetation, such as chaparral in southern California, often causes significant damage to infrastructure and ecosystems. A physically-based semi-empirical model to predict surface fire spread rate is used in the United States to assist in a variety of fire management operations. However, the applicability of this model to shrub fuel beds such as live chaparral is currently unknown. In order to study wildfire spread in living vegetation, four of the most common brush fuels in southern California, chamise (Adenostoma fasciculatum), manzanita (Arctostaphylos densiflora), scrub oak (Quercus berberidifolia) and ceanothus (Ceanothus sp.), were burned and compared. The observed fire behavior included flame height, mass loss rate and temperature structure. Measurements of these properties were obtained from circular live brush fuel beds of various diameters. Simple explicit relationships were established to describe the fire plume. From digital video pictures of the fire plume, it was observed that flame height increased mainly with heat release rate. The temperature structure of the fire plume was measured by an infrared camera. By using successive images of the temperature field, a recently developed thermal particle image velocity (TPIV) algorithm was applied to estimate flow velocities in the vicinity of the flame. Results are in broad agreement with other experimental data obtained over gas and liquid fuels.