Broad availability of molecular sequence data allows construction of phylogenetic trees with 1000s or even 10 000s of taxa. This paper reviews methodological, technological and empirical issues raised in phylogenetic inference at this scale. Numerous algorithmic and computational challenges have been identified surrounding the core problem of reconstructing large trees accurately from sequence data, but many other obstacles, both upstream and downstream of this step, are less well understood. Before phylogenetic analysis, data must be generated de novo or extracted from existing databases, compiled into blocks of homologous data with controlled properties, aligned, examined for the presence of gene duplications or other kinds of complicating factors, and finally, combined with other evidence via supermatrix or supertree approaches. After phylogenetic analysis, confidence assessments are usually reported, along with other kinds of annotations, such as clade names, or annotations requiring additional inference procedures, such as trait evolution or divergence time estimates. Prospects for partial automation of large-tree construction are also discussed, as well as risks associated with ‘outsourcing’ phylogenetic inference beyond the systematics community.