Reported here is direct imaging (and diffraction) by using 4D ultrafast electron microscopy (UEM) with combined spatial and temporal resolutions. In the first phase of UEM, it was possible to obtain snapshot images by using timed, single-electron packets; each packet is free of space-charge effects. Here, we demonstrate the ability to obtain sequences of snapshots (''movies'') with atomic-scale spatial resolution and ultrashort temporal resolution. Specifically, it is shown that ultrafast metal-insulator phase transitions can be studied with these achieved spatial and temporal resolutions. The diffraction (atomic scale) and images (nanometer scale) we obtained manifest the structural phase transition with its characteristic hysteresis, and the time scale involved (100 fs) is now studied by directly monitoring coordinates of the atoms themselves.imaging ͉ diffraction ͉ electron crystallography ͉ vanadium dioxide E lectron microscopy has become a pivotal tool in numerous fields of study, from materials to biological imaging. In a previous publication (1), we introduced the concept of singleelectron imaging for the development of 4D ultrafast electron microscopy (UEM). The potential of UEM was demonstrated by obtaining images and diffraction patterns of materials and biological cells (1), and the scope of future applications has been highlighted in recent commentaries and reviews (refs. 2 and 3; see also ref. 4 and references therein). Because single-electron packets have no significant space-charge broadening, images and diffraction patterns are observed with atomic-scale spatial resolution and with the time resolution being fundamentally determined by the ultrashort duration of the optical pulse introduced to generate the photoelectrons in the microscope.The images and diffraction patterns reported (1) were ''snapshots'' at a particular point in time. However, as shown in Fig. 1, by delaying a second initiating optical pulse to arrive at the sample in the microscope with controlled time steps, it is possible to obtain a series of such snapshots with a well defined frame time. Unlike optical pump-probe experiments, this experimental task, for the microscope, is technically nontrivial for a number of reasons. To determine the zero time point, the clocking of the electron packet and optical pulse at the sample must be made with femtosecond time precision. Moreover, in contrast to these all-optical experiments, the cross-correlation between electron and photon pulses requires a new methodology. In addition, for 120-keV electrons, the group velocity of electron packets in the microscope is two-thirds the speed of light, and care has to be taken to account for this group velocity mismatch. Overcoming these hurdles in conjunction with attaining high quality, nanometer-scale samples in the microscope provides the capability of observing the dynamical changes of systems in the far-fromequilibrium state with the combined resolutions mentioned above.With the 4D UEM arrangement shown in Fig. 1, we demonstrate such studies of...