Being able to perform computational estimations efficiently is an important skill. Furthermore, computational estimation experiments are used to study general principles in strategy development. Rounding strategies are common in computational estimation. However, little is known about whether and when children use a mixed-rounding strategy (i.e., both rounding up and down in one estimation) and how demanding this is in comparison to only rounding-down or only rounding-up. Therefore, we systematically varied the size of unit digits (i.e., the rightmost digit in a whole number) in 72 addition problems. These estimation problems were presented to fourth graders. Most children preferred to use mixed-rounding on mixed-unit problems and therefore adjusted their strategy choice to the individual unit digits in a calculation. Additionally, the sum of units barely influenced children's strategy choice. On mixed-rounding calculations, the proportion of best strategy use was comparable to that of rounding-up and the latencies to produce an estimate with mixed-rounding were between those for rounding-down and rounding-up. Therefore, the mixed-rounding strategy was in the difficulty range of the two more frequently studied rounding strategies; it was also the preferred strategy for mixed-unit problems by children who adapted their estimation strategies. Based on these findings we argue that research into strategy development with estimation tasks should also include mixed-rounding to improve ecological validity