In the area of reverse engineering, the normal of point cloud is the basis of data processing such as smoothing, simplifying and fusing. The consistent adjustment of normal orientation (or called normal adjustment in short) is an essential step for normal estimation. In this paper, a new algorithm for normal adjustment is proposed. It mainly researches on how to get the right adjustment effect and how to make it faster. To adjust the normal rightly, the angle between the original normals of two points is used as the basis of spread order. A suitable exploration mode, in which only the border of adjusted points and unadjusted is explored, is chosen to make the adjustment process efficient. To get more efficiency, the spread based on threshold value is used to adjust several points for one time. The spread based on minimum value is also used. The latter mode is not as efficient as the former one, but it ensures that the spread will not suspend. Experiments result indicates that the normal adjustment algorithm is efficient.