Exponential growth in numerical series and graphs is grossly underestimated in an intuitive extrapolation task. Subjects’ extrapolations are well described by a model with two parameters only: one for underestimation of the nonlinear growth, the other for linear compensation. The size of the effect is considerable; it is not unusual that two-thirds of the subjects produce estimates below 10% of the normative value. The effect increases with the exponent of the stimulus series, and with addition of a constant to the stimulus series. Neither special instructions about the nature of exponential growth nor daily experience with growth processes enhanced the extrapolations.