The recent process analytical technology (PAT) initiative has put an increased focus on online sensors to generate process-relevant information in real time. Specifically for fermentation, however, introduction of online sensors is often far from straightforward, and online measurement of biomass is one of the best examples. The purpose of this study was therefore to compare the performance of various online biomass sensors, and secondly to demonstrate their use in early development of a filamentous cultivation process. Eight Streptomyces coelicolor fed-batch cultivations were run as part of process development in which the pH, the feeding strategy, and the medium composition were varied. The cultivations were monitored in situ using multi-wavelength fluorescence (MWF) spectroscopy, scanning dielectric (DE) spectroscopy, and turbidity measurements. In addition, we logged all of the classical cultivation data, such as the carbon dioxide evolution rate (CER) and the concentration of dissolved oxygen. Prediction models for the biomass concentrations were estimated on the basis of the individual sensors and on combinations of the sensors. The results showed that the more advanced sensors based on MWF and scanning DE spectroscopy did not offer any advantages over the simpler sensors based on dual frequency DE spectroscopy, turbidity, and CER measurements for prediction of biomass concentration. By combining CER, DE spectroscopy, and turbidity measurements, the prediction error was reduced to 1.5 g/l, corresponding to 6% of the covered biomass range. Moreover, by using multiple sensors it was possible to check the quality of the individual predictions and switch between the sensors in real time.