Typical parameter estimation procedures using inversion codes include 1) construction of a hydraulic model, 2) collection of quantitative data such as drawdown and flow rate. Quantitative and qualitative data that describe the spatial variability and geological features (soft data) introduce geological complexity to the model construction, whereas drawdown and flow data (hard data) are used to constrain the inversion process. Using a series of data sets generated in a three-dimensional synthetic aquifer, we investigated how the accuracy of parameter estimation improved with increasing quantities of soft data incorporated into the model construction. The synthetic aquifer (208.3 (L) x 116.8 (W) x 57.2 cm (H), constructed using five well-characterized sands) has a moderately heterogeneous stationary spatially correlated random field with an embedded lens-like layer of fine sand, resulting in a composite aquifer that is non-stationary. The pressure distribution within the aquifer was measured at 92 locations in the three-dimensional aquifer. Parameter estimation was performed using observed drawdowns and different quantities of soft data. The results showed that the accuracy in parameter estimation improves with the quantities of soft data incorporated into the model. The improvement was drastic when information on heterogeneity was incorporated. This suggests that, for reducing the estimation errors, a calibration model needs to be carefully constructed not only with distinct geological features but also with some level of heterogeneity at smaller scale.