This study investigated the removal effectiveness of Escherichia coli by five pilot bioretention units with different vegetations. The bioretention units were originally planted or seeded with one of the following vegetation: shrubs, grass species specified for Texas highway applications, Texas native grasses, Bermudagrass (Cynodon dactylon), and none (control unit). To mimic highway conditions, the vegetation units were allowed to develop naturally outdoors and received no weed control. Longest hydraulic retention time (HRT) was observed in the control unit. Compared to the control unit, the vegetation caused 50% reduction of HRT in the shrub unit and 70-90% reduction of HRTs in the three grass units. Removal efficiency of E. coil of each unit, from high to low, was: control unit (97%) > the shrub unit (88%) > the unit originally seeded with Bermudagrass (76%) > the unit originally seeded with Texas highway grasses (57%) > the unit originally seeded with Texas native grasses (48%). Presence of common weeds, Giant Ragweed and Johnson grasses, might be responsible for the low E. coli removals observed in the grass units. The results provide fundamental and needed knowledge for designing bioretention that can effectively remove pathogens from highway stormwater runoff in semi-arid regions like Texas. (C) 2011 Elsevier B.V. All rights reserved.