Forced-gradient tracer tests (FGTTs) and uniform flow tracer tests (UFTTs) were conducted in a unique physically and chemically highly heterogeneous three-dimensional test aquifer to study the differences in the scale dependence and spatial variability of dispersivity and retardation factors estimated from temporal moments of breakthrough curves due to changes in the flow configuration and source sizes. It is experimentally shown that the distance scales at which dispersivities from FGTTs approach a constant value are larger than those from UFTTs. These discrepancies are largely emphasized when dispersivities from small-source FGTTs are compared with those from large-source UFTTs. Remarkably, point source FGTTs showed a late but substantial scale dependence of dispersivity. In practice, at usual field test scales (few spatial correlation scales), dispersivities estimated from small source FGTTs will significantly underestimate those needed to make predictions on the fate and transport of tracer plumes under natural gradients. Nonetheless, at large distances, asymptotic dispersivities from large source FGTTs were similar to those from large source UFTTs. A negative correlation between R and lnK developed an increase in dispersivities for reactive tracers in both uniform and forced-gradient flow systems. Whereas in uniform flows retardation factors showed a decreasing scale dependence with distance approaching the arithmetic average of the retardation factor field at large distances, retardation factors from FGTTs exhibited a strong increasing scale dependence. Spatial variability of dispersivities and retardation factors estimated from FGTTs were larger than those from UFTTs.