• Casella & Berger, "Statistical Inference"
          An introductory grad-level textbook on major concepts.
• Gelman, Carlin, Stern & Rubin, "Bayesian Data Analysis"
          A practical and well-written guide to Bayesian analysis. No deep philosophy with mindshaking Bayesian-vs-frequentist fights.
• Haykin, "Neural Networks, a Comprehenisve Foundation"
          A really comprehensive review of general machine-learning concepts, neural networks, radial basis function and kernel methods, principal component analysis, support vector machines, boosting methods, decision trees, entropy and mutual information approaches, and stochastic machines. The last 4 of 15 chapters cover topics on neurodynamics; I do not find these particularly useful at the moment. The previous 11 chapters, however, are very relevant for HEP data analysis and recommended for consumption by every physicist. This book does a really good job of explaining the underlying math - one does not have to be a professional mathematician to understand it.
• Webb, "Statistical Pattern Recognition"
          There is a substantial overlap between this book and Haykin's "Neural Networks." Webb offers additional topics in non-parametric density estimation and discusses linear (and quadratic) discriminant analysis, projection pursuit, multivariate adaptive regression splines, feature selection and clustering algorithms. Appendix A includes various measures of dissimilarity that I found particularly amusing in connection with the recent spark of curiosity to multidimensional goodness-of-fit measures. This book covers even more relevant material than Haykin's; the disadvantage is that the discussion is very succinct, and it may be not so easy to digest certain topics without prior background.
• Hastie, Tibshirani & Friedman, "The Elements of Statistical Learning"
          My favorite modern introduction to machine learning methods. A brief yet very informative overview of most methods and issues.
• Haupt & Haupt, "Practical Genetic Algorithms"
          Genetic algorithms are not covered in any other machine learning book mentioned here. This is a good overview of the basics.
• Efron & Tibshirani, "An Introduction to the Bootstrap"
          Bootstrap (and jackknife as a specific subclass) is a method for evaluating bias and variance of an estimator by resampling available data. This method should be considered for HEP analysis whenever it is hard or impossible to obtain a new sample, for example, Monte Carlo is too expensive or there is no more data available.
• D'Agostino & Stephens, "Goodness-of-Fit Techniques"
          An extensive (and somewhat exhaustive) review of GOF techniques, mostly for univariate problems.