Regional groundwater flow was observed and characterized in a laboratory flow tank containing homogeneous and isotropic sand. The tank which consists of plexiglass material is 1.56 m long, 0.1 m wide, and 0.9 m high, and is equipped with seven piezometer nests with which hydraulic heads could be measured directly. Both horizontal and unit-basin flow were observed in the different experiments. Conceptual models of the flow tank were constructed based on the observed data and numerical modeling was performed using two simulation codes, Flowpath and Flonet/trans. In the horizontal flow case, equipotentials generated by the two codes were similar. However, the modeled equipotentials were different from observed equipotentials especially on the inflow side of the model domain. A sudden drop in groundwater heads that was observed as water first entered into the tank could not be reproduced by simulation, thus the model appeared to be calibrated on the outflow end but not on the inflow end. This statement also appears to be true for the unit-basin flow case. This analysis demonstrates how insufficient head data may lead to wrong conclusions during model calibration.