Fine-tuning from a collection of models pre-trained on different domains (a\n"model zoo") is emerging as a technique to improve test accuracy in the\nlow-data regime. However, model selection, i.e. how to pre-select the right\nmodel to fine-tune from a model zoo without performing any training, remains an\nopen topic. We use a linearized framework to approximate fine-tuning, and\nintroduce two new baselines for model selection -- Label-Gradient and\nLabel-Feature Correlation. Since all model selection algorithms in the\nliterature have been tested on different use-cases and never compared directly,\nwe introduce a new comprehensive benchmark for model selection comprising of:\ni) A model zoo of single and multi-domain models, and ii) Many target tasks.\nOur benchmark highlights accuracy gain with model zoo compared to fine-tuning\nImagenet models. We show our model selection baseline can select optimal models\nto fine-tune in few selections and has the highest ranking correlation to\nfine-tuning accuracy compared to existing algorithms.\n