In the European Investigation into Cancer and Nutrition study (EPIC), physical activity (PA) has been indexed as a cross-tabulation between PA at work and recreational activity. As the proportion of non-working participants increases, other categorization strategies are needed. Therefore, our aim was to develop a valid PA index for this population, which will also be able to express PA continuously. In the German EPIC centers Potsdam and Heidelberg, a clustered sample of 3,766 participants was re-invited to the study center. 1,615 participants agreed to participate and 1,344 participants were finally included in this study. PA was measured by questionnaires on defined activities and a 7-day combined heart rate and acceleration sensor. In a training sample of 433 participants, the Improved Physical Activity Index (IPAI) was developed. Its performance was evaluated in a validation sample of 911 participants and compared with the Cambridge Index and the Total PA Index. The IPAI consists of items covering five areas including PA at work, sport, cycling, television viewing, and computer use. The correlations of the IPAI with accelerometer counts in the training and validation sample ranged r = 0.40-0.43 and with physical activity energy expenditure (PAEE) r = 0.33-0.40 and were higher than for the Cambridge Index and the Total PA Index previously applied in EPIC. In non-working participants the IPAI showed higher correlations than the Cambridge Index and the Total PA Index, with r = 0.34 for accelerometer counts and r = 0.29 for PAEE. In conclusion, we developed a valid physical activity index which is able to express PA continuously as well as to categorize participants according to their PA level. In populations with increasing rates of non-working people the performance of the IPAI is better than the established indices used in EPIC.