Although, modern Optical Character Recognition (OCR) models perform efficiently on clean or mildly degraded printed text images, they still struggle significantly when processing images with incomplete or heavily eroded characters and words. This is a common issue that occurs when performing OCR on scanned newspapers and old books, where textual details are often lost due to: (i) typesetting imperfections, (ii) gradual fading of printed ink, and (iii) natural deterioration of paper quality. However, the absence of a dedicated benchmarking dataset presents a significant challenge for developing a versatile OCR solution capable of effectively tackling the above issue. To address this challenge, we introduce EroPT, a benchmarking dataset consisting of 2,437 manually annotated images of eroded printed text. Further, we evaluate cutting-edge OCR models to benchmark their performance in the EroPT dataset for accurate recognition and extraction of textual information. Our results reveal a notable performance decline in state-of-the-art OCR models when handling severely eroded samples. Fine-tuning on the EroPT dataset results in significant accuracy gains, achieving up to a decrease of 24.92 in Character Error Rate (CER) and 33.66 in Word Error Rate (WER) metrics. Despite these improvements, the recognition accuracy remains suboptimal, indicating room for further enhancement. These findings highlight the inherent limitations of existing OCR methods in processing images with eroded printed text thus making our dataset a valuable benchmark for evaluating future OCR models in this specific task.