A combined experimental and computational study on a rectangular lobed mixer is performed. A series of simulations based on a steady Reynolds-averaged Navier-Stokes Simulation (RANS) are conducted to analyse the mixing mechanisms of large-scale streamwise structure shed by the trailing edge of lobed mixer, with emphasis being placed on the effect of turbulence modeling and inflow conditions. The simulations are validated in respect of velocity and scalar distribution against the data obtained through Particle Image Velocimetry (PIV) and Nanoparticle-based Planar Laser Scattering (NPLS) technique. The computational results predicted by the SST k -ω turbulence model show better agreement with the experimental data. But the small-scale turbulence structures are not captured accurately by these turbulence models. The convoluted shear layer shed from trailing edge is stretched and rotated by the large-scale streamwise vortices, forming an unstable 'pinching-off' structure, which increases the interfacial area. And at the interface of two streams, a large number of small-scale turbulence structures are formed, which contribute a lot to the mixing enhancement along with the increased interfacial area. The streamwise vorticity decays more rapidly with the decrease of velocity ratio and total pressure ratio of two streams. The scalar thickness which reflects the mixing rate of two streams increases with the decreasing velocity ratio and total pressure ratio.