The role of pH in the fermentative H-2 production from an upflow acidogenic granule-based reactor was investigated in this study. Experimental results show that all H-2 partial pressure, H-2 production rate and H-2 yield were pH-dependent, in the range of 2.8 x 10(4)-5.2 x 10(4) Pa, 61-145 ml-H-2 l(-1) h(-1) and 0.68-1.61 mol-H-2 mol-glucose-1, respectively. The maximum H-2 partial pressure was observed at pH 3.4, while both maximum H-2 production rate and H-2 yield were found at pH 4.2. Acetate, propionate, butyrate, i-butyrate, valerate, caporate and ethanol were present in the effluent of this UASB reactor, and their distribution was also pH-dependent. As pH was decreased from 4.2 to a lower level of 3.4 or increased to a higher level of 6.3, the fermentative type of this H-2-producing reactor would shift from butyrate-type to caporate- or ethanol-type. Thermodynamic analysis was performed to explore the possible metabolic pathways of caproate and valerate formation. The metabolic pathway of caproate formation was pH-dependent, while that of valerate formation was pH-independent. A neural network model was designed, trained and validated. It was able to successfully describe the daily variations of H-2 partial pressure and H-2 Yield of the reactor, and to predict its steady state performance at various pHs. (c) 2005 Elsevier Ltd. All rights reserved.