We study the problem of high-dimensional variable selection with some good ini tial estimator which is element-wise estimation consistent but not variable selection consistent.We show that by using the nonnegative garotte procedure, we can ob tain a final estimator which is both efficient in estimation and consistent in variable selection.Our setting is high-dimensional, i.e.we allow the number of variables increase almost as fast as exp(n).We also study the conditions under which the ridge estimator can be a good initial estimator.Numerical studies are conducted to support the theory.