The measurement or estimation of bioprocess states is critical for process control and optimization applications. However, certain disturbances or unknown inputs can generate significant model-plant mismatch if not considered by the process models. To better estimate the process states in the presence of these disturbances or unknown inputs, a nonlinear unknown input observer is applied. Experimental studies of batch and fed-batch operations of a bioreactor were performed using a recombinant Saccharomyces cerevisiae to produce beta-carotene. Previously developed kinetic models produce model-plant mismatch with changes to the initial conditions or operating mode of the bioreactor. The observer is applied to the bioreactor system to estimate the batch and fed -batch state variables. State estimates from the designed observer are compared to model predictions and experimental measurements. Results show improved state estimation over first-principles model predictions when applying the unknown input observer to the nonlinear dynamic process with unknown disturbances.