In general, a tomogram cannot be observed immediately after the acquisition of a series of specimen tilt images, but is instead observed after the post-processing of the tilt series alignment, which often requires a substantial amount of time. Moreover, for general specimens, the automatic acquisition of the tilt series is difficult because field-of-view tracking frequently fails as the tilt angle or specimen thickness increases.In this study, we focus on the improvement of the field-of-view autotracking technique for the purpose of online tomography reconstruction and propose a new alternative technique [1,2]. The method we proposed uses a so-called 'back-projected ray image' instead of a specimen tilt image. The back-projected ray image is a cross-section image calculated from each projection image only during reconstruction. As a result of a study on 'ray images', the quality and accuracy of the cross-correlation between a pair of neighboring ray images among the tilt series were observed to be very high compared with those between a pair of projection images. We observed that a back projected ray image reliably cross-correlates with other neighboring ray images at the position of an existing three-dimensional object. The proposed method can therefore consistently track the field-of-view, overcoming the weakness of a conventional image-matching-based method. In addition, the present method is simple, and high speed processing is expected to be achieved because fast Fourier transform (FFT) and inverse fast Fourier transform (IFFT) algorithms can be used.We applied this method to real specimens in online experiments using a TEM and thereby demonstrated its successful performance. Online autotracking experiments with thin-section samples were used to demonstrate the effectiveness of the proposed method. The field-of-view was automatically tracked with high accuracy through a tilt angle range. Furthermore, online tomograms were obtained immediately after the last specimen tilting. With increases in the tracking speed, in situ tomographic observations for analyzing the dynamic behavior might become feasible in the future.jmicro;63/suppl_1/i23-a/DFU058F1F1DFU058F1Fig. 1.Comparison of the proposed autotracking method with the conventional PCF based alignment method using the yeast cell thin-section. a and b: Reconstructed X-Y cross-section images from tracking results at 8° increment angle with the PCF method and with the proposed method. N, nucleus; V, vacuole; NVJ, nucleus-vacuole junction. c: A reconstructed cross-section image from autotracking result at 1° increment angle with the proposed method. (scale bar: 100 nm).