The security of organizational databases has received considerable attentio
n in the literature in recent years. This can be attributed to a simultaneo
us increase in the amount of data being stored in databases, the analysis o
f such data, and the desire to protect confidential data. Data perturbation
methods are often used to protect confidential, numerical data from unauth
orized queries while providing maximum access and accurate information to l
egitimate queries. To provide accurate information, it is desirable that pe
rturbation does not result in a change in relationships between attributes.
In the presence of nonconfidential attributes, existing methods will resul
t in such a change. This study describes a new method (General Additive Dat
a Perturbation) that does not change relationships between attributes. Al e
xisting methods of additive data perturbation are shown to be special cases
of this method. When the database has a multivariate normal distribution,
the new method provides maximum security and minimum bias. For nonnormal da
tabases, the new method provides better security and bias performance than
the multiplicative data perturbation method.