For ill-defined systems (like environmental and economical systems) with sparse and uncertain data, Hornberger and Spear have proposed, as a variation on Monte Carlo simulation analysis, the so-called regionalized sensitivity analysis (RSA) to determine dominant model parameters. In this paper a minor modification to the RSA is proposed to mitigate the effect of arbitrarily chosen initial parameter intervals. This modification concerns the application of a parameter space identification method prior to the RSA to offer more reliable (with respect to the observed systems' behavior) initial parameter intervals. An alternative procedure to the RSA is also proposed. Herein, the results from a parameter-space identification method are analyzed directly in terms of eigenvectors (principal axes) and eigenvalues characterizing the associated identified parametric subspace. These methods are applied to a simple water-quality model with sufficient as well as sparse data. Comparison with the results from period-average analysis and maximum-likelihood estimation reveals that consistent information is obtained.