A defaunation experiment mimicking the effects of a harmful algal bloom (HAB) on benthic macroinvertebrate assemblages (> 500 µm) was conducted at a hydrodynamically active soft-substrate site in Wellington Harbour, New Zealand, to test the recovery rate (return to predisturbance state) of temperate benthic macroinvertebrate assemblages and to elucidate the main factors influencing the recovery process. Tarpaulins were used to create anoxic conditions by smothering the sediment for 65 d. Assemblage recovery in treatment plots was studied for 1 yr and compared with assemblage composition in undisturbed adjacent control plots. Recovery was slow until Day 70, at which time abundance of individuals (N) and number of species (S) increased synchronously in treatments and controls. Within 10 mo, univariate indices (N, S and also species diversity H' and evenness J') of treatment assemblages returned to values observed for the control assemblages. Multivariate analyses showed that fluctuations in assemblage composition were most pronounced in the first 100 d in treatment replicates, but decreased thereafter as the recovering assemblage became more similar to the undisturbed one. However, after 1 yr, even though the composition of treatment and control assemblages was converging, differences in composition were still significant. Based on the observed trajectory of recovery, complete assemblage recovery is predicted to take approximately 2 yr. Timing of the disturbance in relation to seasonal recruitment events was identified as an important factor for assemblage recovery.