Turbulent mixing in confined shear flows, including a rectangular jet and a rectangular wake, was investigated using both non-simultaneous particle image velocimetry (PIV) and planar laser-induced fluorescence (PLIF) and simultaneous PIV and PLIF techniques. Based on the hydraulic diameter and the bulk velocity, the Reynolds number was 50,000 for the jet, and was 37,500 for the wake flow. The Schimdt number of the fluorescence dye is 1,250;The experimental data were analyzed for flow statistics such as mean velocity, Reynolds stresses; turbulent kinetic energy, turbulent dissipation, concentration mean and variance, turbulent fluxes, turbulent viscosity and diffusivity, and turbulent Schmidt number. The constraint of the flows resulted in some different characteristics of the profiles of the above flow properties from those in free shear flows, indicating that the sidewalls affect the evolution of the flow significantly. Furthermore, the mean concentration gradient was found not aligned with the turbulent flux vector, therefore the simple gradient transport model using a scalar turbulent diffusivity cannot represent the behavior of the turbulent flux in such flows;Spatial correlation of turbulent fluxes and concentration were also studied. For the jet, the Ru'o' was elliptical in shape with a major axis tilted downward with respect to the streamwise axis, whereas the Rv'o' was a horizontally oriented ellipse. The Ro'o' was an ellipse with the major axis inclined due to the mean shear in the flow. For the wake, however, there were a positive and a negative R u'o' field, symmetric around the basis point. The behavior of large-scale structures in both the velocity and concentration field was studied using linear stochastic estimation with a defined event of concentration fluctuation. The streamwise growth of the structure size increased linearly initially but then grew more slowly;Finally, linear models for the conditional velocity and the conditional scalar were tested against the experimental data and gave good predictions. A gradient PDF model for the conditional velocity was found to give poor predictions. However, the improved gradient PDF model predicted velocities much better.