Flow meters are commonly used in manholes to monitor the flow rate for sewer operation and management. However, the large-scale deployment of flow meters in a sewer system is cost-prohibitive due to their high costs and the need for frequent maintenance. This paper proposes a soft sensor that estimates flow rates based on water level measurements in a manhole. The soft sensor is powered by a Multilayer Perceptron (MLP) model, of which the input is a time series of measured water levels, and the output is the corresponding flow rate. The model was trained using flow and water level data collected from three real-life manholes as well as data generated for a simulated manhole using a three-dimensional Computational Fluid Dynamics model. In all cases, the trained MLP produced satisfactory estimations of the flow data in the test phase. Further studies of the simulated manhole showed that the soft sensor is robust against noise and bias associated with the water level and flow measurements. The proposed soft sensor provides a potentially cheaper and more durable alternative to traditional flow meters for sewer applications.