The influence of the type of soybean derivatives as nitrogen sources, as well as the simultaneous influence of the concentrations of nitrogen and carbon sources in the production of clavulanic acid (CA), by Streptomyces clavuligerus, were investigated. Firstly, two runs in shake flasks were performed utilizing soybean flour (SF) and soybean protein isolated (SPI) in the culture medium with concentration of 1.6 g L-1 total nitrogen (TN). The CA production in the culture medium with SF was much higher, about double the production obtained with SPI. SF was utilized in the additional experiments to study the quantitative influences of the concentrations of SF as nitrogen source and soybean oil (SO) as carbon source, on CA production. Six runs were performed in a bench scale bioreactor. The experiments had the following concentrations: TN (1.6, 2.4 and 3.2 g L-1) and SO (16.0 and 23.0 g L-1), respectively. The cellular growth, evaluated in terms of theological parameter consistency index (K) of the broths, reached a maximum value (K-max proportional to the initial concentration of total nitrogen, but K-max was not influenced by the initial concentration of soybean oil (C-SO). In general, it has been observed that the consistency index (K) decreased rapidly due to the cell fragmentation caused by high volumetric power input. In the range studied, CA production increased with the decrease of C-SO and the increase of C-SF. The batch cultivation, utilizing lipid as supplemental substrate, simulated a fed-batch cultivation, in which the glycerol feeding rate is defined by the lipid hydrolysis rate, and CA production was favored by a slower supply of glycerol and fatty acids. Maximum CA production of 906 mg L-1 was obtained with C-SF = 40.0 g L-1 and Cso = 16.0 g L-1, respectively. So far, this is the highest CA production ever found in the literature using wild strain of S. clavuligerus in batch cultivations. An empirical relationship was proposed to correlate maximum CA concentration with CSF and Cso. The correlation obtained fitted the experimental data well, allowing the prediction of CA production for different experimental conditions and further process optimization. (c) 2006 Elsevier Inc. All rights reserved.