When an array of closely spaced sensors is applied, even if the number of sensors is very large, the error magnification phenomenon will render some cross spectra of wave records useless to the estimates of directional spectra because these cross spectra contain about the same information as others. Consequently, only few cospectra or quadrature (quad) spectra can provide truly useful information about the directional distribution when the ordinary spectral analysis procedures are applied. In this paper the sources of certain magnified errors are identified and several error-reducing strategies are developed so that three more cospectra or quad spectra become useful to the estimates of directional spectra when a compact array with three closely spaced biaxial current meters plus a wave staff is utilized. While most of the strategies are directly relevant to reduction of the random errors, the influences of other data acquisition and processing errors are also fully taken into account. These error-reducing strategies can improve the directional resolution of all estimators. Their effectiveness is illustrated through a series of model tests with the truncated Fourier series method (TFSM) and the Long and Hasselmann (1979) and Herbers and Guza (1990) minimum nastiness methods. In the TFSM analysis the additional three cospectra or quad spectra allow determination of the third and fourth angular harmonics of directional spectra, after which even the fifth and sixth harmonics can be determined approximately by a simple iterative scheme in most cases. For the modern Long and Hasselmann and Herbers and Guza techniques, the improvements by the present strategies are a bit marginal when the nonnegativity constraint of these techniques can very effectively optimize the estimates, but this may depend on the array geometry, and in some cases there is a clear improvement. ¿ American Geophysical Union 1995