Spatial patterns of gene expression span many scales, and are shaped by both local (e.g. cell-cell interactions) and global (e.g. tissue, organ) context. However, most in situ methods for profiling gene expression either average local contexts or are restricted to limited fields of view. Here we introduce sci-Space, a scale-flexible method for spatial transcriptomics that retains single cell resolution while simultaneously capturing heterogeneity at larger scales. As a proof-of-concept, we apply sci-Space to the developing mouse embryo, capturing the approximate spatial coordinates of profiled cells from whole embryo serial sections. We identify genes including Hox-family transcription factors expressed in an anatomically patterned manner across excitatory neurons and other cell types. We also show that sci-Space can resolve the differential contribution of cell types to signalling molecules exhibiting spatially heterogeneous expression. Finally, we develop and apply a new statistical approach for quantifying the contribution of spatial context to variation in gene expression within cell types.