Valuation models lie at the core of both financial theory and practice, yet we lack systematic evidence on how professionals value assets, which models perform best, and why. To make progress on these questions, we analyze valuation models in 1.1 million equity analyst reports. While, on average, simpler multiples-based models generate more accurate forecasts than more complex discounted cash flow (DCF) models, this masks important heterogeneity: skilled analysts produce superior forecasts with DCF models, especially for hard-to-value firms, underscoring the importance of expertise when employing complex models. To establish that model-specific expertise matters, we exploit a quasi-exogenous shock that forced some analysts to switch valuation models, and show that their forecast accuracy subsequently declines relative to analysts with established experience using the new approach. This highlights a fundamental trade-off between simplicity and sophistication in valuation, where optimal method choice depends on analyst characteristics, such as skill. Finally, given their unconditional superior performance, we study how analysts determine multiples. Analysts use historical, current, and peer-based reference points to contextualize their choice of multiples, but they do not use these benchmarks mechanically when determining their prices. Moreover, we show that sensitivity analyses have become increasingly common, bull and bear scenarios are asymmetric and account for greater downside risk, and their inclusion is associated with more conservative forecasts.