Traditionally, oxygen consumption (VO₂), which reflects workload of physically demanding jobs, has been estimated from heart rate (HR) using the linear relationship between both variables. However, due to the presence of external factors, such as fatigue, emotional stress and fitness level, this relationship becomes nonlinear, especially at low workload intensity. This study presents a new method to estimate oxygen consumption from heart rates using adaptive neuro-fuzzy inference system (ANFIS), which is capable of handling uncertainties and nonlinearity. In a laboratory experiment, eight participants performed two step-tests in consecutive days during which oxygen consumption and heart rate were measured. Data from step-test 1 were used to develop individual ANFIS for each participant. The individual ANFIS were then tested and compared with traditional linear models using the dataset obtained from step-test 2. The results indicated increase in VO2 estimation accuracy of 38% (at low workload intensity, HR<90 bpm) and 21% (in general, throughout HR range). Individual ANFIS show potential to replace linear models at workplaces with small working population or when accurate estimation of physical workload is desired.