Arsenic (As) removal in subsurface-flow constructed wetlands (CW) planting with vetiver grasses was experimented by comparing between two different configurations; (i) deep-bed units (dpCW) with length to depth (L:D) ratio = 2 and (ii) shallow-bed units (shCW) with L:D ratio = 8; operating at hydraulic retention time (HRT) of 6, 9, and 12 days. The tracer study of CW units revealed that no effect of L:D ratio on dispersion number could be determined, but affecting to the effective volume ratio. Based on the data obtained from the pilot-scale experiments of CW units for 117 days, it is apparent that the dpCW could achieve relatively high As removals (52.9%, 59.2%, and 72.1% at HRT of 6, 9, and 12 days, respectively). Analysis of As mass balance showed that only 0.2–0.4% of As input was uptaken by vetiver grasses whereas the major portion was retained in the CW media (38.9–77.6%). Forms of the retained As was determined by sequential fractionation which could indicate As complexation with iron and manganese on the media surface of 31–38% and As trapping into the media of 42–52% of the total. No obvious difference of As fractions in bed of between dpCW and shCW units was observable.