We aimed at showing that children are able to identify and draw a line of best fit through a cloud of points according to the concept of linear regression. A sample of 72 children in third and fourth primary school grades were given two different tasks. In a first task, 10 different sheets with three illustrations of the same scatterplot with different lines of regression to choose from were presented. In a second task, scatterplots were given and children were asked to draw a line of best fit with pencil and ruler. In the first task bisector regression was preferred (χ2 = 15.21, df = 4, p = .004), whereas in the second conventional regression ordinary least squares (X|Y) was favored (χ2 = 27.14, df = 4, p < .001). In summary, the preference of the traditional regression model using a minimization in the vertical dimension was only partly supported by our data.