PropertiesConverter makes some bad assumptions about the Strings that appear in the name-value pairs. From the PropertiesConverter javadoc: Special converter for java.util.Properties that stores properties in a more compact form than java.util.Map. Because all entries of a Properties instance are Strings, a single element is used for each property with two attributes; one for key and one for value. The problem is that the XML spec (http://www.w3.org/TR/REC-xml/#AVNormalize) and the org.xmlpull.mxp1.MXParser "normalize" attributes to remove tabs, newlines, and carriage returns. So a value inside a Properties object called "my<tab>property" gets normalized to "my<space>property" when reading the attribute (see MXParser.parseAttribute()). When writing Properties to a file, Java escapes the tabs with a \t so that it would appear in the property files as key=my\tproperty PropertiesConverter should probably be removed entirely and Properties should be encoded just like any other Map using XML entities. It could still use attributes, but it would have to use [[CDATA]] in order to work with the XML spec.