Since the introduction of contrast-enhanced MR angiography (CE-MRA) (1), the technique has evolved to achieve higher image quality in terms of spatial resolution and signal-to-noise ratio (SNR). Recently, much effort has been directed at trying to increase temporal resolution so that a multiphase acquisition will always have an uncontaminated arterial phase volume, eliminating the need for contrast bolus timing. Further, time-resolved CE-MRA, if fast enough, can follow the first pass of contrast through the vasculature to show flow dynamics, which is necessary in certain clinical scenarios such as arteriovenous shunting, retrograde flow, or delayed vessel enhancement. Currently, x-ray digital subtraction angiography (DSA) (2) is the clinical standard for time-resolved angiography. However, unlike MRI, x-ray DSA utilizes ionizing radiation and nephrotoxic contrast material and carries the small but definite risk of introducing embolic material. In addition, MRI can provide true 3D imaging, not just projection imaging, in order to visualize the complex 3D geometry of blood vessels. The combination of 3D and time-resolved imaging is also known as “4D” imaging. CE-MRA would be a viable option for time-resolved angiography if the SNR, spatial resolution, and temporal resolution were sufficient for the clinical problem. However, there is an inherent trade-off between these three parameters in MRI, given a fixed field-of-view (FOV). In recent times, the research community has developed a great number of novel acceleration, or undersampling, techniques to increase spatial and temporal resolution at the expense of SNR. These techniques can be broadly classified as temporal and/or spatial undersampling. Temporal undersampling involves the estimation of missing k-space samples from past and future acquired samples at the same spatial frequency. The “sliding window” technique was among the first acceleration strategies to be developed (3,4). It involves the repeated acquisition of N segments of a whole volume such that frames can be formed by any contiguous set of N segments. The frame rate is increased by N, but the data are low-pass filtered in the temporal dimension. The “keyhole” technique first acquires a high-resolution volume, and subsequently only updates the low-frequency information, where most of the image lies (5). BRISK (6), TRICKS (7), and STBB analysis/acquisition (8) techniques built on previous ideas by acquiring the central k-space lines at a higher sampling rate than peripheral lines. Spatial undersampling techniques involve the estimation of missing k-space lines from other acquired lines at different spatial frequencies in the same time frame. One technique, in use for many years now, involves the asymmetric sampling of k-space. Spatial frequencies higher than a certain threshold are not acquired on one side of k-space, and the missing data are either zero-filled or extrapolated. This concept can be extended by applying it in multiple dimensions for large overall undersampling factors. Parallel imaging is comprised of a group of techniques which use multiple receiver coils to speed up acquisition. The SMASH technique was one of the first described (9), and it has matured into GRAPPA (10), which is presently state-of-the-art. Another class of parallel imaging approaches is based on SENSE (11). Alternative k-space trajectories may have their own specific undersampling techniques. For example, the radial trajectory can be angularly undersampled (12,13). Combinations of temporal and spatial undersampling, such as PR-TRICKS (14), k-t SENSE (15), and TSENSE (16) can be used, and may even be preferable to a single acceleration technique alone. Another concern in developing a time-resolved CE-MRA sequence relates to ordering the acquisition of k-space lines. Since lines are acquired in a serial fashion, the passage of contrast causes a modulation of the signal according to the order in which the lines are acquired (17). Better ordering methods, such as elliptical centric ordering (18), will minimize the artifact associated with serial measurement. Subtraction for CE-MRA is conventionally performed using a single time frame as the mask. The mask is chosen at a time prior to the arrival of contrast. For timed CE-MRA, the acquisition typically consists of three volumes: the mask, arterial phase, and venous phase. Time-resolved acquisitions open up a new set of applications and, therefore, the way in which subtraction is performed may need to be reconsidered. In the context of time-resolved imaging, this article attempts to address these issues of undersampling, ordering, and subtraction with a CE-MRA technique combining a 3D radial trajectory (cylindrical sampling), sliding window reconstruction, pseudorandom ordering, and a sliding subtraction mask. The goals are to demonstrate with simulations and healthy volunteer studies that 1) for the sliding window reconstruction, the artifact associated with the radial trajectory is less severe than that for the spin-warp trajectory (Cartesian sampling); 2) proper ordering is essential when combining the radial trajectory with the sliding window method; and 3) the sliding mask subtraction technique better depicts the wash-in and wash-out of contrast than conventional subtraction.