In this paper, I first argue that similarity accounts of scientific pictures fail with more realistic cases of scientific pictures. My primary case study is the picture of a black hole, from which I develop an interpretation-based account of picture representation analogous to how models represent: a picture represents a designated target system iff, once interpreted, it exemplifies properties that are imputed to the target via a de-idealising function. Then, I argue that the justification of the inferences from mechanically produced pictures depends on their causal mechanisms of production, in contrast with the standard justificatory strategies employed for model inferences.