Segmenting brain tumors from MRI scans is crucial in neuroimaging analysis, aiding in diagnosis, treatment planning, and disease monitoring. This project introduces Dual Path Attentive Feature Network (DPAFNet), an innovative deep learning architecture for precise brain tumor segmentation irrespective of size and appearance. DPAFNet has an encoder-decoder block inspired by the UNet architecture, integrating selective dual skip connections: one with a multi-resolution module, an attention gate integrated to capture spatial dependencies and contextual information from multi-modal MRI data, and another from the same encoder layer concatenated with the attention gate output. With approximately 2.35 million parameters, less than the UNet and modified UNet architectures, the model is computationally efficient and suitable for real-time analysis. Assessments conducted on BraTS2020 reveal higher segmentation accuracy compared to baseline networks. DPAFNet achieves state-of-the-art performance, with Dice scores of 0.9003, 0.9301, and 0.8854 for non-tumor core(NTC), whole tumor(WT), and enhancing tumor(ET), along with corresponding Hausdorff distances of 1.75, 2.8, and 1.36.