I just want to know how does the table resulting from readXML look like, say if the xml file looks like this:

<item attr="some attribute">
<descirption>anything</description>
</item>

I can reference tables directly by the Tables collection like this:

ds.ReadXml(xml);
... ds.Tables[i]

then I can access rows and columns using the rows collection:

ds.Tables[3].Rows[i].ItemArray(j);

but how can I access an "attribute" of any xml node?

It's in the Row. I fixed your xml a bit :)

var xml = "<item attr="some attribute"><description>anything</description></item>";
var ds = new DataSet();

ds.ReadXml( new StringReader( xml ), XmlReadMode.Auto );

var ia = ds.Tables[0].Rows[0].ItemArray;
var att = ia[1]; // att == "some attribute"

If you don't have a schema, you might have to check the column to determine what it is.

Per comment: You will see I am letting it infer the schema (XmlReadMode.Auto). It takes elements under the root node as Rows then adds the attributes in order and then the value in the element. So for example the following XML ...

var xml = "<items>
             <item attr1='attr1' attr2='attr2'>
                 <description>desc1</description>
             </item>
             <item attr1='attr3' attr2='attr4'>
                 <description>desc2</description></item>
           </items>";

I will get two rows (one for each item) with Columns for attr1, attr2 and description. You can change the way it interprets the XML using a schema.