The study aims at assessing in long-term trials a gravity-driven ultrafi ltration pilot plant designed for a capacity of 5 m3 d−1. The unit was operated in South Africa with Ogunjini surface water and was run with restricted chemical intervention or maintenance (no backfl ush, no aeration, no crossfl ow and no chemical). Under South African environmental conditions and with direct fi ltration of the river water and only one manual drainage of the membrane reactor every weekday, the unit could fulfi l the design specifi cation in terms of water production (5 m3 d−1) as long as the turbidity of the raw water remained in a reasonable level (up to 160 NTU), with a fi ltration fl ux typically 4–6 l h m−2 (corrected at 20°C). This value was in the same range as the lab results and was consistent with the fi rst phase results (around 5–7 l h−1 m−2 after biosand fi ltration). However, the fl ux dropped signifi cantly to a range of 2–4 l h−1 m−2 after a rain event resulting in a turbidity peak over several days up to >600 NTU. This demonstrated that for variable raw water types with expected turbidity peaks above 100 NTU, a pre-treatment would be required for the system (biosand fi lter or other). The performance of microbiological tests confi rmed the integrity of the membrane and the ability of the system to achieve advanced disinfection.