This paper deals with the designs of robust observers for uncertain models, with application to the activated sludge process. We assume that the model is such that the growth rate is unknown, and the measurements and inflow are noisy. However we suppose that their upper and lower bounds are known. Under appropriate hypotheses, we are able to build interval observers giving dynamic bounds containing the variables to estimate. Besides, if we know a priori the probability density of every single uncertain parameter, we can synthetize interval observers with confidence level in order to estimate the probability density of the state. We apply these tools to the sludge process and compare the results to probabilistic simulation.