In a companion paper we have reported a $>5\sigma$ detection of degree scale $B $-mode polarization at 150 GHz by the BICEP2 experiment. Here we provide a detailed study of potential instrumental systematic contamination to that measurement. We focus extensively on spurious polarization that can potentially arise from beam imperfections. We present a heuristic classification of beam imperfections according to their symmetries and uniformities, and discuss how resulting contamination adds or cancels in maps that combine observations made at multiple orientations of the telescope about its boresight axis. We introduce a technique, which we call "deprojection", for filtering the leading order beam-induced contamination from time ordered data, and show that it removes power from BICEP2's $BB$ spectrum consistent with predictions using high signal-to-noise beam shape measurements. We detail the simulation pipeline that we use to directly simulate instrumental systematics and the calibration data used as input to that pipeline. Finally, we present the constraints on $BB$ contamination from individual sources of potential systematics. We find that systematics contribute $BB$ power that is a factor $\sim10\times$ below BICEP2's 3-year statistical uncertainty, and negligible compared to the observed $BB$ signal. The contribution to the best-fit tensor/scalar ratio is at a level equivalent to $r=(3-6)\times10^{-3}$.

We report on the design and performance of the BICEP2 instrument and on its three-year data set. BICEP2 was designed to measure the polarization of the cosmic microwave background (CMB) on angular scales of 1 to 5 degrees ($\ell$=40-200), near the expected peak of the B-mode polarization signature of primordial gravitational waves from cosmic inflation. Measuring B-modes requires dramatic improvements in sensitivity combined with exquisite control of systematics. The BICEP2 telescope observed from the South Pole with a 26~cm aperture and cold, on-axis, refractive optics. BICEP2 also adopted a new detector design in which beam-defining slot antenna arrays couple to transition-edge sensor (TES) bolometers, all fabricated on a common substrate. The antenna-coupled TES detectors supported scalable fabrication and multiplexed readout that allowed BICEP2 to achieve a high detector count of 500 bolometers at 150 GHz, giving unprecedented sensitivity to B-modes at degree angular scales. After optimization of detector and readout parameters, BICEP2 achieved an instrument noise-equivalent temperature of 15.8 $\mu$K sqrt(s). The full data set reached Stokes Q and U map depths of 87.2 nK in square-degree pixels (5.2 $\mu$K arcmin) over an effective area of 384 square degrees within a 1000 square degree field. These are the deepest CMB polarization maps at degree angular scales to date. The power spectrum analysis presented in a companion paper has resulted in a significant detection of B-mode polarization at degree scales.