Abstract In semiconductor-device inspection using scanning electron microscopes (SEMs), the irradiation dose of the electron beam becomes lower because of increasing needs for higher throughput and lower damage to the samples. Therefore, it is necessary to form images using fewer primary electrons, making noise reduction of SEM images one of the main challenges. We have modeled the imaging process of SEMs, which consists of the generation of primary, secondary and tertiary electrons (PEs, SEs and TEs, respectively), and detection. Furthermore, a method to accurately evaluate the fluctuation in the number of SEs and TEs are proposed. We found that SEM-image noise can be minimized by directly detecting SEs generated in the sample, in which case the fluctuation in the number of SEs determines the image quality. The variance number of SEs emitted from a 500-eV PE irradiation onto a Si wafer is 1.9 times as large as the value derived assuming a Poisson process. A Monte-Carlo simulation result was used to explain the experimental results and predict that PE energy less than 1 keV suppresses the fluctuation in the number of SEs, and consequently, the SEM-image noise level. These findings provide a method for determining imaging conditions that improve the throughput of SEMs.