Number line estimation tasks are frequently used to learn about numerical thinking, learning, and development. These tasks are often interpreted as though estimates are determined by overall magnitudes of target numerals, rather than specific instantiating digits. Yet estimates are strongly biased by leftmost digits. For example, numbers like "698" are placed too far to the left of numbers like "701" on a 0-1,000 line. This "left digit effect" or "left digit bias" has been investigated little in children, and only on electronic tasks. Here, we ask whether left digit bias appears in paper-and-pencil estimates, and whether it differs for paper-based versus computer-based tasks. In Study 1, 5- to 8-year-old children completed a 0-100 number line task on paper. In Study 2, 7- to 11-year-olds completed a 0-1,000 paper task. In Study 3, adults completed tasks on paper in both ranges. Large left digit effects were observed for children aged 8 years or older and adults, but we did not find evidence for left digit bias in younger children. Study 4 compared paper and computer tasks for adults and children aged 9-12 years. Strong left digit bias was observed in all conditions, with a larger effect for the paper-based task in children. Large left digit effects in number line estimation emerge regardless of task format, with a developmental trajectory broadly consistent with other studies. For children in the age range that reliably exhibits left digit bias (but not adults), paper-and-pencil number line estimation tasks elicit even greater bias than computer-based tasks.