Sensor network research is still in its infancy. Few real systems are deployed and little experimental data from sensor networks is available to test proposed protocol designs. Due to lack of experimental data and sophisticatedmodels derived fromsuch data, most data processing algorithms from the sensor network literature are evaluated with data generated from simple parametric models. We identify a few widely-studied classes of problems that are potentially sensitive to data input: Statistics estimation of the field data; Data compression; and Field estimation. We use them as examples to investigate the dependency of algorithm performance on data. For each class of problem, given the selected problem and algorithm instance, we systematically study how the algorithm performance varies across a range of data input. We also demonstrate how different data input can change the algorithm performance dramatically, the performance comparison between two algorithms may even change depending on the different data inputs. In the end, we propose our synthetic data generation framework and recommend evaluating algorithms across a wide range of data input.