Model revision is a challenging task in image classification. It is usually performed to correct the predictions of misclassified examples. Support examples, which are similar to the misclassified example and which either carry the opposite class label (near misses) or the same class label (near hits) can serve as a basis for shifting decision boundaries. Although, near miss and near hit explanations provide the means for exploring and revising a model’s decision boundary with the help of example-based XAI and through minimal changes, they are not yet sufficiently explored for correcting misclassifications. We propose and illustrate two approaches utilizing near misses and near hits for model revision in binary image classification: a similarity-based method that generates near misses and near hits in the vicinity of a misclassified sample and a concept-based method that suppresses a minimal number of neural concepts present in a misclassified example while keeping necessary neural concepts that are present in near hit or near miss examples.