In the current paper, we develop an approach to early science pedagogy that is based on insights about how complex adaptive systems function. Complexity approaches have an important advantage over traditional information-processing approaches: They anticipate the proverbial 'mind with a mind of its own' without having to postulate exclusively mental constructs. They also offer insights about key determinants of learning and effective pedagogy, again without postulating exclusively mental constructs. For complex adaptive systems, learning depends on the presence of sufficiently salient novelty (i.e., variability), and it depends on the presence of sufficiently salient repetitions or ordered patterns (i.e., stability). Science learning, therefore, requires science-relevant novelty and science-relevant patterns of order. Equipped with these insights, we address two challenges of early science pedagogy: (1) how to combine children's self-guided explorations with teachers' strategic interventions, and (2) how to minimize the chances of generating misconceptions about science. The answer lies in creating a learning context that maximizes science-relevant variability and science-relevant stability. If both aspects are abundantly available, a child's self-guided explorations are effective. Conversely, if either aspect is missing, efforts must be made to add them strategically to a child's experience. Adding science-relevant stability is particularly challenging, yet crucial to avoid science misconceptions.