Based on a non-isotropic form of error correlation function derived for radar radial-wind analysis on conic surfaces of radar scans, a new statistic method is developed to estimate Doppler wind observation error covariance and retrieved background wind error covariance from Doppler wind innovations. The innovations are obtained by subtracting independently derived background winds (projected on each radar beam at each range gate) from Doppler observed winds. In this study, the background vector winds are provided by the real-time Doppler wind retrieval system (http://gaussian.gcn.ou.edu:8080/rtime.shtm) applied to the Oklahoma KTLX radar, while independent Doppler wind observations are from the OKC airport TDWR radar. The new method is applied to quality-controlled innovation radial-wind fields collected for the period from 06:01 to 18:57 on 10/26/02. The results are very encouraging and will be reported at the conference.