Noise generated by low Mach number flow in duct networks is important in many industrial applications. In the automotive industry the two most important are the ventilation duct network and the engine exhaust system. Traditionally, design is made based on rule-of thumb or slightly better by simple semi-empirical scaling laws for flow noise. In many cases, strong curvatures and local deviations from circular cross-sections are created due to outer geometry restrictions. This can result in local relatively high flow velocities and complex flow separation patterns and as a result, rule-of thumb and scaling law methods can become highly inaccurate and uncertain. More advanced techniques based on time domain modelling of the fluid dynamics equations together with acoustic analogies can offer a better understanding of the local noise generation, the propagation and interaction with the rest of the system. This investigation focuses on validating an SNGR numerical model to predict flow noise generation due to separation in a circular duct with a 90-degree bend carrying a flow lower than 0.3 Mach number. Experimental results are presented and compared to numerical simulations, based on a combination of steady computational fluid dynamics and the stochastic acoustic analogy by Lighthill, as well as semi-empirical models based two-ports.