This analysis focuses on predicting New York Times (NYT) bestselling fiction books, specifically on the Combined Print & E-Book Fiction list, using book descriptions and book covers. Approximately ten years of data from the NYT Books API and Open Library API is collected, and an exploratory data analysis is performed. All machine learning models are evaluated using performance metrics in the classification report, including accuracy, weighted average precision, and weighted average recall. In addition, the model performances are assessed using the precision-recall curve, ROC curve, and confusion matrix. Although the dataset is fairly balanced, NYT bestsellers are rare in reality, so the classification threshold is tuned for each model and precision is prioritized. Seven text-based models are implemented by utilizing cleaned book descriptions as input, which are transformed using either TF-IDF or BERT. Based on the overall performances, the text-based models with the best performances are Logistic Regression with BERT and XGBoost with TF-IDF, with strong accuracies at 92% and 91%, respectively. Next, five image-based models are implemented by extracting image features from book covers. The overall performances of the image-based models indicated that the XGBoost model with CLIP embeddings and the ResNet50 model are the strongest, with accuracies 87% and 83%, respectively. The best text-based and image-based models are then considered for multimodal modeling. Four multimodal models are implemented and compared. The Logistic Regression model with BERT & CLIP features was the top-performing multimodal model, achieving a strong accuracy of 97%.