3D lane detection plays an integral role in autonomous driving, traffic planning, and intelligent transportation systems. It bolsters driving safety and efficiency, improves navigation accuracy, and aids in comprehending and predicting complex road conditions. However, due to the absence of depth information, monocular 3D lane detection is a challenging task. One common approach is to convert front-view (FV) images or features into bird’s-eye-view (BEV) space using inverse perspective mapping (IPM) and detect lanes based on BEV features. But the reliance of IPM on the assumption of a flat ground and the loss of contextual information hinder accurate 3D reconstruction from BEV representations. Although previous methods based on 3D anchor have attempted to overcome it, they often suffer from the complexity of two-stage processing and encounter difficulties when pooling anchors in a 2D setting. In this paper, we propose a 3D feature amplifier that incorporates crucial height information to shape the spatial characteristics of lanes. Additionally, we redefine more suitable anchors to further exploit the potential of this module. Extensive experiments on two popular 3D lane detection benchmarks demonstrate that our 3DLaneFormer outperforms previous anchor-based methods and BEV-based methods, could achieves state-of-the-art performance within these categories of methods.