The fertilised mammalian egg gives rise to seemingly equivalent blastomeres until the fourth cleavage division, when the first indication of lineage specification appears. At this point, certain blastomeres divide symmetrically and others asymmetrically. When do these apparently identical cells diverge and how do these differences arise? To answer this question, Enkui Duan and colleagues performed single-cell transcriptional analysis of human and mouse blastomeres (p. 3468). By studying the mammalian zygote, in which transcription – a known source of heterogeneity during mitosis – is mostly silent, the authors showed that small biases in gene expression arise after the first cleavage division from the unequal distribution of cellular substances between daughter cells, called ‘partitioning errors’. These are especially pronounced for transcripts present in small quantities, which are more likely to be asymmetrically distributed. As cleavage divisions progress, the activation of embryonic transcription minimizes or amplifies the initial biases through positive or negative feedback regulation. Furthermore, the authors show that lineage specification is driven by the relative ratio of pairs of competing lineage specifiers, such as Cdx2 and Carm1, the levels of which are determined by both cleavage history and de novo transcription. This study shows that symmetry breaking leading to lineage specification is a continuous process that emerges as early as the two-cell stage, before morphological differences between blastomeres are detectable.