Multi-modal large language models (MLLMs) have demonstrated significant advancements in tasks related to image-text generation. However, MLLMs may generate outputs that contradict visual inputs, a phenomenon known as the hallucination within MLLMs. A comprehensive and objective assessment of hallucinations in MLLMs is crucial for advancing their further development. To this end, we introduce the Hallucination Primary Examination (HoPE) of Multi-modal Large Language Models, a benchmark for MLLMs hallucination assessment. HoPE offers a holistic evaluation of object-level, attribute-level, and relationship-level hallucinations in MLLMs through the utilization of Scholastic Assessment Tests (SAT) quiz-like formats, including judgmental, multiple-choice, and fill-in-the-blank questions. We utilize different evaluation metrics for different question types and propose HoPE scores to measure the overall performance of the model. Leveraging the HoPE, we deliver a more scientific and objective assessment of hallucinations in widely used MLLMs. We also thoroughly analyze the capabilities and weaknesses of MLLMs on the basis of the examination results and have made recommendations.