This paper describes and explains some regularities in the frequency of numbers in text. An analysis of number frequencies in text corpora in Dutch, English, German, and French confirms the expectation that frequency is highly dependent on two factors: magnitude and roundness. Roundness (defined as number frequency in an approximation context) proves to be related to three arithmetical properties: 'lO-ness', '2-ness', and '5-ness'. In predicting the frequency of numbers irrespective of their context '2Y2-ness' should be added to these factors, as is suggested in the work of Sigurd (1988). The role of the four number characteristics found in this study can be explained by the preference of the language user for using base numbers, and for doubling and halving quantities.