Duplicate bugs pose a significant challenge that consumes substantial resources and can complicate the bug triage process, requiring extra work to identify and merge duplicates. Several automated duplicate bug detection methods use Natural Language Processing to handle this problem. Bug reports are often long and contain multiple sections that can show some textual (dis)similarities. This disparity may affect the duplicate bug detection process and hence results in inefficient resource utilization. In this work, we study the impact of bug report sections on the detection of duplicates especially when these sections show some (dis)similarities. Filtering out the most pertinent sections can greatly alleviate computational load and reduce the chances of overlooking potential duplicate bugs. Using less sections would also reduce the cost of the duplicate bug detection system as less tokens may be used. To achieve our objective, we developed and analyzed two types of models. One section-based models are used to analyze the individual impact of bug report sections, whereas cross section-based models are used to analyze their collective impact. These models leverage a siamese network constructed from pretrained DistilRoBERTa [1] and fine-tuned for classification by the Multi-Layer Perceptron (MLP). Our findings reveal that the “title” and “description” sections show the highest relevance in duplicate bug detection, achieving f1-scores of 98.93% and 98.29% respectively. Conversely, the “steps to reproduce” and “actual results” sections tend to cause confusion when distinguishing between duplicate reports, which often results in a high misclassification rate.