The use of the two-stage least squares (2 SLS) procedure for estimating nonrecursive social science models is often impractical when multiple feedback linkages are required. This is because 2 SLS is extremely sensitive to multicollinearity. The standard statistical solution to the multicollinearity problem is a biased, variance reduced procedure known as ridge regression (RR). However, RR is a stochastic solution when it should be nonstochastic, and does not meet the boundary condition; thus, a new ridge technique called normalization ridge regression (NR) is proposed. When NR is used in an analogous fashion to 2 SLS it should theoretically provide a superior solution to multiple feedback estimating problems than either 2 SLS or 2 SRR. All three techniques (2 SLS, 2 SRR, and 2 SNR) are used to estimate a multiple feedback model of school science achievement for a national probability sample of Finnish 14 year-olds. The 2 SNR estimates were compared to both 2 SLS and 2 SRR estimates. In general, the 2 SNR results were the most consistent and confirmed the posited advantages of the biased, variance-reduced 2 SNR procedure over the other nonrecursive estimating methods. (Author/DWH)