Interaction-free measurement (IFM) has been proposed as a method of reduced-damage electron microscopy [1][2][3]. Recently, with the implementation of Mach-Zehnder interferometers in conventional transmission electron microscopes (TEMs), it has become possible to potentially implement IFM in these tools [4][5][6]. Therefore, a comparison of the theoretical performance of IFM with conventional microscopy is of interest [7].In this work, we theoretically analyzed the performance of IFM imaging of both opaque-and-transparent and semitransparent samples, and compared it to the performance of conventional scanning transmission electron microscopy (STEM) [8]. For opaque-and-transparent samples, we compared the performances of the two schemes using two metrics -"## , the probability of misidentifying an opaque pixel as transparent or vice-versa, and %&'&(" , the mean number of electrons required to image an opaque pixel. Figure 1(a) compares "## for IFM with that for conventional STEM, at a constant %&'&(" of 2.5 electrons per pixel, for (the prior probability of a given pixel being opaque) between 0 and 1. We performed this comparison for IFM and conventional STEM both with and without a detector for scattered electrons ( , ), to account for different microscope configurations. We can see that "## was lower for IFM (green dashed-dotted curve) than conventional STEM (purple solid curve) for a wide range of . This includes the important limit of low , which is commonly encountered for high-transparency electron microscopy samples.In figure 1(b), we compare "## vs %&'&(" for IFM and STEM, for = 0.5. In these calculations, we included a sample re-illumination scheme based on updating a prior for each pixel of the sample after each round of illumination with a Poisson-limited electron beam, based on the statistics at the imaging detectors. The re-illumination for a pixel ceases once a stopping criterion is met. This scheme reduces %&'&(" for both IFM and STEM imaging to their ideal values -⅔ for IFM imaging with , (green solid curve with square markers) and 1 for STEM imaging with , (purple solid curve with circle markers). Therefore, conditional re-illumination allowed us to circumvent the Poisson statistics of the beam.For semi-transparent samples, we treated the transparency ∈ [0,1] as a continuous random variable. The statistics at the imaging detectors can be used to form an estimate of , and the performance of the estimator can be analyzed by looking at its mean squared error (MSE). For unbiased estimators, the inverse of the classical Fisher Information (FI) forms a lower bound for this MSE (Cramér-Rao bound). We found that the FI for IFM and STEM imaging was identical, shown by the solid blue curve in figure 2. Figure 2 also shows the MSE for two estimators for -7 and 8 , calculated using Monte-Carlo simulations. These estimators use the counts from the imaging detectors in different ways -7 (purple dashed curve) averages over these counts to estimate , while 8 (orange dashed-dotted curve) uses the square of the dif...