Purpose: To show that open‐field normalization prevents a common error in measurements of the modulation transfer function (MTF) caused by zero‐frequency normalization. Method and Materials: Models describing zero‐frequency and open‐field normalization are used to derive the resulting measured MTF obtained from a finite region of interest (ROI) of an edge image in terms of the true system MTF. Simulated edge images containing Gaussian blur, as well as images of a tungsten sheet acquired on a GE flat‐panel detector in clinical use were used to calculate the MTF. Measurements were made using both zero‐frequency and open‐field normalization with ROIs ranging in size from 1–15 cm × 10 cm and 1–15 cm × 5 cm. Results: Use of a finite ROI results in truncation of the system line‐spread function (LSF) causing the zero‐frequency value of the measured MTF to be less than the true MTF value of unity. Subsequent zero‐frequency normalization results in inflation of the MTF values at all non‐zero frequencies. Data from the clinical system shows a 5% inflation of values for an ROI of 10 cm, increasing to 14% for an ROI of 1 cm. Open‐field normalization accurately determines MTF values at all frequencies away from zero frequency. The open‐field normalized zero‐frequency value is equal to the area of the truncated system LSF, and is less than unity. Conclusion: Open‐field normalization measurements provide a good estimate of the true MTF if the zero‐frequency value is disregarded with the understanding that it is expected to have a value less than unity. This work should be used in MTF‐measurement guidelines and recommendations on acceptable ROI sizes and normalization techniques.