<p>Accurate estimation of the seismic demand of buildings requires carefully considering both the ground motion group size (<i>n</i>) and the angle of seismic input (ASI). Existing studies focus more on the relative effect of either <i>n</i> or ASI on evaluation results. There is ongoing debate on the <i>n</i> and ASIs required in deterministic and probabilistic seismic performance assessments, and no specific guidelines have been established for engineering practice. To address this concern, this study proposes a framework for assessing statistical bias of response and providing engineering-oriented recommendations for <i>n</i> and ASI in nonlinear response history analysis of mid-rise buildings. Firstly, a regrouping procedure is presented to generate 100 suites of ground motions with a specified smaller size of <i>n</i> from a benchmark large dataset of ground motions. Next, three statistical indices are defined to precisely characterize the bias between the small-sample and benchmark large-sample ground motions, namely <i>bias</i>(<i>µ</i>), <i>sdev</i>(<i>µ</i>), and <i>bias</i>(<i>β</i>), which quantifies the effects of <i>n</i> and ASI on errors of the estimated central tendency and variability of response in deterministic and probabilistic analyses. Based on an acceptable statistical bias, the required values of <i>n</i> and ASI for different conditions are recommended. Finally, to facilitate practical application, a simple yet efficient model of the correction coefficient (<i>κ</i>) of variability is proposed to save computational time and improve assessment accuracy. Results show that at least 10 pairs of records and 4 input directions are required to ensure reliable estimates of expected seismic response for mid-rise buildings. The bias in variability can be adjusted using the <i>κ</i>. These engineering-practical recommendations contribute to resolving the ongoing debate regarding the necessary <i>n</i> and ASI in seismic performance assessments of mid-rise buildings.</p>