Recently, graph neural networks (GNNs) have achieved remarkable performances for quantum mechanical problems. However, a graph convolution can only cover a localized region, and cannot capture long-range interactions of atoms. This behavior is contrary to theoretical interatomic potentials, which is a fundamental limitation of the spatial based GNNs. In this work, we propose a novel attention-based framework for molecular property prediction tasks. We represent a molecular conformation as a discrete atomic sequence combined by atom-atom distance attributes, named Geometry-aware Transformer (GeoT). In particular, we adopt a Transformer architecture, which has been widely used for sequential data. Our proposed model trains sequential representations of molecular graphs based on globally constructed attentions, maintaining all spatial arrangements of atom pairs. Our method does not suffer from cost intensive computations, such as angle calculations. The experimental results on several public benchmarks and visualization maps verified that keeping the long-range interatomic attributes can significantly improve the model predictability.