A number of applications require the precise tracking or position estimation of an object unresolved in the system optics. This paper evaluates several (NxN) centroid-like interpolation algorithms (N=2,3,4,5) designed to make these estimates to subpixel accuracy. Analytic and Monte Carlo results are presented. The tracking sensor examined was a staring mosaic array (100% coverage assumed) of detectors assumed to be device-noise (e.g., CCD noise) limited. The detector size was varied parametrically to determine the relative performance and to obtain the optimum configuration. The optics blur spot was assumed Gaussian. The sources of error considered to affect the algorithm performance were the systematic algorithm bias (or positional error), the random noise (or jitter error), and the postcalibration residual detector responsivity nonuniformities. The results were applied to the design of the SIRTF Fine Guidance Sensor. Track accuracy improves with signal-to-noise ratio (SNR), until limited by algorithm inaccuracies or focal-plane nonuniformity. But blur spot distortion has significant impact on algorithm performance. Among the algorithms tested, the relative SNR performance improved as N decreased. However, extreme sensitivity to algorithm bias error limited the use of the (2x2) algorithm to cases with positional requirements z L/25 (even with correction). The (3x3) algorithm is then optimum for positional requirements z, L/100 (with correction). Higher (NxN) algo-rithms are required for greater positional accuracy.