It is shown that the so-called cluster method for determining the horizontal gradients of the horizontal velocity has some basic limitations caused by restrictive physical assumptions. For large-area clusters there is aliasing of these estimates by large-scale shear, while for small-area clusters, the displacement of the statistical cluster center from a physically significant flow center can produce seriously aliased cluster model parameter estimates. These effects are independent of typical experimental problems such as measurement error and statistical reliability. An alternative to the cluster paradigm is proposed. This, however, requires solving a nonlinear inverse problem to recover estimates of the horizontal velocity gradients. ¿ American Geophysical Union 1988