SpatialGlue is a graph neural network model with a dual-attention mechanism that integrates spatial and omics data to decipher spatial domains in tissues. It combines intra-omics integration of spatial location and omics measurements with cross-omics integration to capture anatomical details and resolve spatial domains more accurately than existing methods. SpatialGlue was tested on data from different tissue types, including spatial epigenome–transcriptome and transcriptome–proteome modalities, and demonstrated superior performance in identifying cell types and resolving spatial structures. It can integrate three modalities and scales well with data size. SpatialGlue uses a dual-attention mechanism to adaptively capture the importance of different modalities, resulting in more accurate integration. It was benchmarked against other methods on simulated and experimental data, showing better performance in quantitative metrics and spatial resolution. SpatialGlue was also applied to mouse brain and spleen data, revealing finer-grained tissue structures and identifying previously unannotated cell types. It outperformed other methods in resolving spatial domains and distinguishing cell subsets. SpatialGlue is a spatially aware method that integrates multiple omics data modalities to provide a holistic view of cellular and tissue properties. It is designed to handle the challenges of integrating spatial multi-omics data for downstream analyses and is applicable to various tissue types and technologies. The model uses attention mechanisms to adaptively learn the relative importance between omics modalities and spatial location, making it suitable for spatial multi-omics analysis. SpatialGlue is a promising tool for analyzing spatial multi-omics data and has the potential to facilitate new biological discoveries.SpatialGlue is a graph neural network model with a dual-attention mechanism that integrates spatial and omics data to decipher spatial domains in tissues. It combines intra-omics integration of spatial location and omics measurements with cross-omics integration to capture anatomical details and resolve spatial domains more accurately than existing methods. SpatialGlue was tested on data from different tissue types, including spatial epigenome–transcriptome and transcriptome–proteome modalities, and demonstrated superior performance in identifying cell types and resolving spatial structures. It can integrate three modalities and scales well with data size. SpatialGlue uses a dual-attention mechanism to adaptively capture the importance of different modalities, resulting in more accurate integration. It was benchmarked against other methods on simulated and experimental data, showing better performance in quantitative metrics and spatial resolution. SpatialGlue was also applied to mouse brain and spleen data, revealing finer-grained tissue structures and identifying previously unannotated cell types. It outperformed other methods in resolving spatial domains and distinguishing cell subsets. SpatialGlue is a spatially aware method that integrates multiple omics data modalities to provide a holistic view of cellular and tissue properties. It is designed to handle the challenges of integrating spatial multi-omics data for downstream analyses and is applicable to various tissue types and technologies. The model uses attention mechanisms to adaptively learn the relative importance between omics modalities and spatial location, making it suitable for spatial multi-omics analysis. SpatialGlue is a promising tool for analyzing spatial multi-omics data and has the potential to facilitate new biological discoveries.