Salary data from a single firm are analyzed in an effort to identify the firm's wage policy. We find that employees are partly shielded against changes in external market conditions; that wage variation within a job level is large both cross-sectionally and for individuals over time, often leading to substantial real wage declines; that wage increases are serially correlated even controlling for observable characteristics; and that promotions and wage growth are strongly related, even though promotion premiums are small relative to the large wage differences between job levels. None of the major theories of wage determination can alone explain the evidence.