Legal Textual Entailment Recognition (LTER) requires sophisticated reasoning to navigate the syntactic complexity and logical structures inherent in legal language. This study introduces a novel structured prompting approach that leverages Large Language Models (LLMs) without resource-intensive fine-tuning. We design specialized prompts with intricate logical structures that guide model inference to align with the patterns of legal reasoning. Through systematic evaluation on COLIEE 2024 and VLSP 2023 benchmarks, our approach achieves 97 % accuracy on Vietnamese legal texts and outperforms GPT-4 and strong previous research on multiple English legal reasoning tasks. These results demonstrate that carefully designed prompting strategies can unlock complex legal logic in LLMs while maintaining computational efficiency.