As PdI as dimensionless, it cannot be defined as Saed Abbasi stated, for it would be a number of nanometers.
The actual definition of PdI according to Malvern (referring directly to the ISO 13321:1996 norm) is based on the Cumulants analysis that is performed to fit the correlation function: Ln(C(tau))=a+b*tau+c*tau², a third order fit where PdI=2c/b².
(see slide 23 in cme.nuk.edu.tw/files/personal_subject/420_0f8d97a2.ppt)
Eventually, in the theoretical case of size distribution made of a single gaussian peak, the PdI may also be computed by sigma²/mean², which Saed Abbasi was rather close to.
(see p.18 in http://ncl.cancer.gov/NCL_Method_NIST-NCL_PCC-1.pdf)