Abstract: A knot placement algorithm is proposed to construct a B-spline curve approximating to a dense and ordered point set.Curvature extreme value points of the curve and two end points of the point set are viewed as seed points to construct a B-spline curve approximating to the seed points by the least-squares method and accomplish feature points' refinement and knot vectors' update by the complexity of the curve segment.This process is repeated to obtain an accurate approximating curve until the maximum error is less than the given error bound.Under the same error bound,experimental results showed that our algorithm can reduce more control points than Park's approach,Piegl's approach and Li's approach.The numbers of control points of the curve is equal to that of the feature points after refinement,and the location of the knots is reasonable.