This study investigates the performance of different stationary and semi-stationary multi-pinhole collimator (MPH) designs with a dual-head L-mode configuration as compared to that of a conventional low-energy high-resolution (LEHR) collimator for myocardial perfusion (MP) single-photon emission computed tomography (SPECT). The target field-of-view for the heart was 16 cm with a target resolution of 1 cm. The design parameters were determined by maximizing the detection efficiency based on the system design constraints. Three stationary designs (L-modes I to III) and semi-stationary designs (S-modes I to III) with four detector positions (i.e., 2 L-mode positions at various angles apart) were evaluated. We used the XCAT phantom with Tc-sestamibi distribution and a three-dimensional analytical MPH/LEHR projector to generate noise-free and noisy projections, which were reconstructed using the maximum-likelihood expectation–maximization algorithm with up to 180 updates. A two-dimensional (2D) region-of-interest with 32 × 32 pixels covering the central axial slice of the left ventricle was analyzed using normalized mean-squared error (NMSE), normalized standard deviation (NSD), and a 2D bias map. The optimized MPH collimator design parameters are 12 pinholes, an acceptance angle of 51°, an aperture size of 4.6 mm, a collimator length of 15.8 cm, and an imaging distance of 18.6 cm. The NMSE of L-mode II (0.326) was the lowest among all stationary designs. S-mode II/40° provided the lowest NMSE result (0.14) among all acquisitions, and was 36.1 % better than LEHR. The NSD-NMSE result showed that S-mode II/40° had the best NMSE-NSD trade-off. The 2D bias map of S-mode II/40° approached that of LEHR. Semi-stationary MPH acquisition substantially improved image quality as compared to those obtained with a full stationary design and a conventional LEHR for MP SPECT.