diff --git a/parse.go b/parse.go
index 89e3776b..0937c2ea 100644
--- a/parse.go
+++ b/parse.go
@@ -64,10 +64,11 @@ func Parse(in io.Reader, bytesToRead int64, frameChan chan *frame.Frame, opts ..
 	}
 
 	for !p.reader.rawReader.IsLimitExhausted() {
-		_, err := p.Next()
+		elem, err := p.Next()
 		if err != nil {
 			return p.dataset, err
 		}
+		p.dataset.Elements = append(p.dataset.Elements, elem)
 	}
 
 	// Close the frameChannel if needed
@@ -115,8 +116,9 @@ func NewParser(in io.Reader, bytesToRead int64, frameChannel chan *frame.Frame,
 	optSet := toParseOptSet(opts...)
 	p := Parser{
 		reader: &reader{
-			rawReader: dicomio.NewReader(bufio.NewReader(in), binary.LittleEndian, bytesToRead),
-			opts:      optSet,
+			rawReader:  dicomio.NewReader(bufio.NewReader(in), binary.LittleEndian, bytesToRead),
+			opts:       optSet,
+			datasetCtx: &Dataset{},
 		},
 		frameChannel: frameChannel,
 	}
@@ -167,7 +169,7 @@ func (p *Parser) Next() (*Element, error) {
 		}
 		return nil, ErrorEndOfDICOM
 	}
-	elem, err := p.reader.readElement(&p.dataset, p.frameChannel)
+	elem, err := p.reader.ReadElement(p.frameChannel)
 	if err != nil {
 		// TODO: tolerate some kinds of errors and continue parsing
 		return nil, err
@@ -186,7 +188,6 @@ func (p *Parser) Next() (*Element, error) {
 		p.reader.rawReader.SetCodingSystem(cs)
 	}
 
-	p.dataset.Elements = append(p.dataset.Elements, elem)
 	return elem, nil
 
 }
diff --git a/read.go b/read.go
index 6e653be5..795ce1dd 100644
--- a/read.go
+++ b/read.go
@@ -41,6 +41,11 @@ var (
 type reader struct {
 	rawReader dicomio.Reader
 	opts      parseOptSet
+	// datasetCtx is the top-level context Dataset holding only elements that
+	// may be needed to parse future elements (e.g. like tag.Rows). See
+	// datasetContextElementPassList for elements that will be automatically
+	// included in this context.
+	datasetCtx *Dataset
 }
 
 func (r *reader) readTag() (*tag.Tag, error) {
@@ -154,7 +159,7 @@ func (r *reader) readHeader() ([]*Element, error) {
 
 	// Must read metadata as LittleEndian explicit VR
 	// Read the length of the metadata elements: (0002,0000) MetaElementGroupLength
-	maybeMetaLen, err := r.readElement(nil, nil)
+	maybeMetaLen, err := r.readElementInternal(nil, nil)
 	if err != nil {
 		return nil, err
 	}
@@ -174,7 +179,7 @@ func (r *reader) readHeader() ([]*Element, error) {
 	}
 	defer r.rawReader.PopLimit()
 	for !r.rawReader.IsLimitExhausted() {
-		elem, err := r.readElement(nil, nil)
+		elem, err := r.readElementInternal(nil, nil)
 		if err != nil {
 			// TODO: see if we can skip over malformed elements somehow
 			return nil, err
@@ -466,7 +471,7 @@ func (r *reader) readSequence(t tag.Tag, vr string, vl uint32, d *Dataset) (Valu
 	seqElements := &Dataset{}
 	if vl == tag.VLUndefinedLength {
 		for {
-			subElement, err := r.readElement(seqElements, nil)
+			subElement, err := r.readElementInternal(seqElements, nil)
 			if err != nil {
 				// Stop reading due to error
 				log.Println("error reading subitem, ", err)
@@ -493,7 +498,7 @@ func (r *reader) readSequence(t tag.Tag, vr string, vl uint32, d *Dataset) (Valu
 			return nil, err
 		}
 		for !r.rawReader.IsLimitExhausted() {
-			subElement, err := r.readElement(seqElements, nil)
+			subElement, err := r.readElementInternal(seqElements, nil)
 			if err != nil {
 				// TODO: option to ignore errors parsing subelements?
 				return nil, err
@@ -519,7 +524,7 @@ func (r *reader) readSequenceItem(t tag.Tag, vr string, vl uint32, d *Dataset) (
 
 	if vl == tag.VLUndefinedLength {
 		for {
-			subElem, err := r.readElement(&seqElements, nil)
+			subElem, err := r.readElementInternal(&seqElements, nil)
 			if err != nil {
 				return nil, err
 			}
@@ -537,7 +542,7 @@ func (r *reader) readSequenceItem(t tag.Tag, vr string, vl uint32, d *Dataset) (
 		}
 
 		for !r.rawReader.IsLimitExhausted() {
-			subElem, err := r.readElement(&seqElements, nil)
+			subElem, err := r.readElementInternal(&seqElements, nil)
 			if err != nil {
 				return nil, err
 			}
@@ -693,17 +698,24 @@ func (r *reader) readInt(t tag.Tag, vr string, vl uint32) (Value, error) {
 	return retVal, err
 }
 
-// readElement reads the next element. If the next element is a sequence element,
-// it may result in a collection of Elements. It takes a pointer to the Dataset of
-// elements read so far, since previously read elements may be needed to parse
+// ReadElement reads the next element. This is the top level function that should
+// be called to read elements. If the next element is a sequence element,
+// it may result in reading a collection of Elements.
+func (r *reader) ReadElement(fc chan<- *frame.Frame) (*Element, error) {
+	return r.readElementInternal(r.datasetCtx, fc)
+}
+
+// readElementInternal reads the next element. If the next element is a sequence element,
+// it may result in reading a collection of Elements. It takes a pointer to the Dataset of
+// context elements, since previously read elements may be needed to parse
 // certain Elements (like native PixelData). If the Dataset is nil, it is
 // treated as an empty Dataset.
-func (r *reader) readElement(d *Dataset, fc chan<- *frame.Frame) (*Element, error) {
+func (r *reader) readElementInternal(datasetCtx *Dataset, fc chan<- *frame.Frame) (*Element, error) {
 	t, err := r.readTag()
 	if err != nil {
 		return nil, err
 	}
-	debug.Logf("readElement: tag: %s", t.String())
+	debug.Logf("readElementInternal: tag: %s", t.String())
 
 	readImplicit := r.rawReader.IsImplicit()
 	if *t == tag.Item {
@@ -715,22 +727,23 @@ func (r *reader) readElement(d *Dataset, fc chan<- *frame.Frame) (*Element, erro
 	if err != nil {
 		return nil, err
 	}
-	debug.Logf("readElement: vr: %s", vr)
+	debug.Logf("readElementInternal: vr: %s", vr)
 
 	vl, err := r.readVL(readImplicit, *t, vr)
 	if err != nil {
 		return nil, err
 	}
-	debug.Logf("readElement: vl: %d", vl)
+	debug.Logf("readElementInternal: vl: %datasetCtx", vl)
 
-	val, err := r.readValue(*t, vr, vl, readImplicit, d, fc)
+	val, err := r.readValue(*t, vr, vl, readImplicit, datasetCtx, fc)
 	if err != nil {
 		log.Println("error reading value ", err)
 		return nil, err
 	}
 
-	return &Element{Tag: *t, ValueRepresentation: tag.GetVRKind(*t, vr), RawValueRepresentation: vr, ValueLength: vl, Value: val}, nil
-
+	elem := &Element{Tag: *t, ValueRepresentation: tag.GetVRKind(*t, vr), RawValueRepresentation: vr, ValueLength: vl, Value: val}
+	addToContextIfNeeded(datasetCtx, elem)
+	return elem, nil
 }
 
 // Read an Item object as raw bytes, useful when parsing encapsulated PixelData.
@@ -789,3 +802,20 @@ func (r *reader) readRawItem(shouldSkip bool) ([]byte, bool, error) {
 func (r *reader) moreToRead() bool {
 	return !r.rawReader.IsLimitExhausted()
 }
+
+// datasetContextElementPassList holds the set of DICOM Element Tags that should
+// be included in the context Dataset. These are elements that may be needed to
+// read downstream elements in the future.
+var datasetContextElementPassList = tag.Tags{
+	&tag.Rows,
+	&tag.Columns,
+	&tag.NumberOfFrames,
+	&tag.BitsAllocated,
+	&tag.SamplesPerPixel,
+}
+
+func addToContextIfNeeded(datasetCtx *Dataset, e *Element) {
+	if datasetContextElementPassList.Contains(&e.Tag) {
+		datasetCtx.Elements = append(datasetCtx.Elements, e)
+	}
+}