diff --git a/DESCRIPTION b/DESCRIPTION
index 48b7985..7e2972f 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,8 +1,8 @@
 Package: SeqArray
 Type: Package
 Title: Big Data Management of Whole-genome Sequence Variant Calls
-Version: 1.21.2
-Date: 2018-06-11
+Version: 1.21.3
+Date: 2018-07-29
 Depends: R (>= 3.3.0), gdsfmt (>= 1.10.1)
 Imports: methods, parallel, IRanges, GenomicRanges, GenomeInfoDb, Biostrings,
         S4Vectors
diff --git a/NEWS b/NEWS
index ac1a480..4c6cfda 100644
--- a/NEWS
+++ b/NEWS
@@ -1,4 +1,4 @@
-CHANGES IN VERSION 1.21.1-1.21.2
+CHANGES IN VERSION 1.21.1-1.21.3
 -------------------------
 
 UTILITIES
@@ -17,6 +17,8 @@ NEW FEATURES
     o a new function `seqCheck()` for checking the data integrity of a SeqArray
       file
 
+    o `seqGDS2SNP()` exports dosage GDS files
+
 BUG FIXES
 
     o `seqVCF2GDS()` and `seqVCF_Header()` are able to import site-only VCF
diff --git a/R/Conversion.R b/R/Conversion.R
index 344df9e..b4c02aa 100644
--- a/R/Conversion.R
+++ b/R/Conversion.R
@@ -320,13 +320,14 @@ seqGDS2VCF <- function(gdsfile, vcf.fn, info.var=NULL, fmt.var=NULL,
 # Convert a SeqArray GDS file to a SNP GDS file
 #
 
-seqGDS2SNP <- function(gdsfile, out.gdsfn,
-    compress.geno="ZIP_RA", compress.annotation="LZMA_RA",
+seqGDS2SNP <- function(gdsfile, out.gdsfn, dosage=FALSE,
+    compress.geno="LZMA_RA", compress.annotation="LZMA_RA",
     optimize=TRUE, verbose=TRUE)
 {
     # check
     stopifnot(is.character(gdsfile) | inherits(gdsfile, "SeqVarGDSClass"))
     stopifnot(is.character(out.gdsfn), length(out.gdsfn)==1L)
+    stopifnot(is.logical(dosage) | is.character(dosage), length(dosage)==1L)
     stopifnot(is.character(compress.geno), length(compress.geno)==1L)
     stopifnot(is.character(compress.annotation), length(compress.annotation)==1L)
     stopifnot(is.logical(optimize), length(optimize)==1L)
@@ -335,7 +336,10 @@ seqGDS2SNP <- function(gdsfile, out.gdsfn,
     if (verbose)
     {
         cat(date(), "\n", sep="")
-        cat("SeqArray GDS to SNP GDS Format:\n")
+        if (isTRUE(dosage) | is.character(dosage))
+            cat("SeqArray GDS to SNP GDS Dosage Format:\n")
+        else
+            cat("SeqArray GDS to SNP GDS Format:\n")
     }
 
     # if it is a file name
@@ -345,12 +349,27 @@ seqGDS2SNP <- function(gdsfile, out.gdsfn,
         on.exit({ seqClose(gdsfile) })
     }
 
+    # dosage variable name
+    if (isTRUE(dosage) | is.character(dosage))
+    {
+        if (isTRUE(dosage))
+            dosage <- "annotation/format/DS"
+        else if (substr(dosage, 1L, 18L) != "annotation/format/")
+            dosage <- paste0("annotation/format/", dosage)
+    }
+
     if (verbose)
     {
         dm <- .seldim(gdsfile)
         cat("    # of samples: ", .pretty(dm[2L]), "\n", sep="")
         cat("    # of variants: ", .pretty(dm[3L]), "\n", sep="")
-        cat("    genotype compression: ", compress.geno, "\n", sep="")
+        if (is.character(dosage))
+        {
+            cat("    dosage compression: ", compress.geno, ", from [",
+                dosage, "]\n", sep="")
+        } else {
+            cat("    genotype compression: ", compress.geno, "\n", sep="")
+        }
         cat("    annotation compression: ", compress.annotation, "\n", sep="")
     }
 
@@ -358,19 +377,16 @@ seqGDS2SNP <- function(gdsfile, out.gdsfn,
     gfile <- createfn.gds(out.gdsfn)
     # close the file at the end
     on.exit({
-        closefn.gds(gfile)
-        if (verbose)
-            cat("Done.\n", date(), "\n", sep="")
-        if (optimize)
-        {
-            if (verbose) cat("Optimize the access efficiency ...\n")
-            cleanup.gds(out.gdsfn, verbose=verbose)
-            if (verbose) cat(date(), "\n", sep="")
-        }
+        if (!is.null(gfile)) closefn.gds(gfile)
     }, add=TRUE)
 
     # add a flag
-    put.attr.gdsn(gfile$root, "FileFormat", "SNP_ARRAY")
+    if (!is.character(dosage))
+    {
+        put.attr.gdsn(gfile$root, "FileFormat", "SNP_ARRAY")
+    } else {
+        put.attr.gdsn(gfile$root, "FileFormat", "IMPUTED_DOSAGE")
+    }
 
     # add "sample.id"
     sampid <- seqGetData(gdsfile, "sample.id")
@@ -402,13 +418,34 @@ seqGDS2SNP <- function(gdsfile, out.gdsfn,
         compress=compress.annotation, closezip=TRUE)
 
     # add "genotype"
-    gGeno <- add.gdsn(gfile, "genotype", storage="bit2",
-        valdim=c(length(sampid), 0L), compress=compress.geno)
-    put.attr.gdsn(gGeno, "sample.order")
+    if (!is.character(dosage))
+    {
+        gGeno <- add.gdsn(gfile, "genotype", storage="bit2",
+            valdim=c(length(sampid), 0L), compress=compress.geno)
+        put.attr.gdsn(gGeno, "sample.order")
+        seqApply(gdsfile, "$dosage", as.is=gGeno, .useraw=TRUE, .progress=verbose,
+            FUN = .cfunction("FC_GDS2SNP"))
+    } else {
+        .cfunction("FC_SetNumSamp")(length(sampid))
+        gGeno <- add.gdsn(gfile, "genotype", storage="float32",
+            valdim=c(length(sampid), 0L), compress=compress.geno)
+        put.attr.gdsn(gGeno, "sample.order")
+        seqApply(gdsfile, dosage, as.is=gGeno, .progress=verbose,
+            FUN = .cfunction("FC_GDS2Dosage"))
+    }
 
-    seqApply(gdsfile, "$dosage", as.is=gGeno, .useraw=TRUE, .progress=verbose,
-        FUN = .cfunction("FC_GDS2SNP"))
     readmode.gdsn(gGeno)
+    closefn.gds(gfile)
+    gfile <- NULL
+    if (verbose)
+        cat("Done.\n", date(), "\n", sep="")
+
+    if (optimize)
+    {
+        if (verbose) cat("Optimize the access efficiency ...\n")
+        cleanup.gds(out.gdsfn, verbose=verbose)
+        if (verbose) cat(date(), "\n", sep="")
+    }
 
     # output
     invisible(normalizePath(out.gdsfn))
diff --git a/man/seqGDS2SNP.Rd b/man/seqGDS2SNP.Rd
index 9930754..70727e6 100644
--- a/man/seqGDS2SNP.Rd
+++ b/man/seqGDS2SNP.Rd
@@ -5,13 +5,16 @@
     Converts a SeqArray GDS file to a SNP GDS file.
 }
 \usage{
-seqGDS2SNP(gdsfile, out.gdsfn, compress.geno="ZIP_RA",
+seqGDS2SNP(gdsfile, out.gdsfn, dosage=FALSE, compress.geno="LZMA_RA",
     compress.annotation="LZMA_RA", optimize=TRUE, verbose=TRUE)
 }
 \arguments{
     \item{gdsfile}{character (GDS file name), or
         a \code{\link{SeqVarGDSClass}} object}
     \item{out.gdsfn}{the file name, output a file of VCF format}
+    \item{dosage}{a logical value, or characters for the variable name of
+        dosage in the SeqArray file; if \code{FALSE} exports genotypes,
+        otherwise exports dosages}
     \item{compress.geno}{the compression method for "genotype"; optional
         values are defined in the function \code{add.gdsn}}
     \item{compress.annotation}{the compression method for the GDS variables,
diff --git a/src/ConvToGDS.cpp b/src/ConvToGDS.cpp
index 819efc6..0fc4829 100755
--- a/src/ConvToGDS.cpp
+++ b/src/ConvToGDS.cpp
@@ -2,7 +2,7 @@
 //
 // ConvToGDS.cpp: format conversion
 //
-// Copyright (C) 2015-2017    Xiuwen Zheng
+// Copyright (C) 2015-2018    Xiuwen Zheng
 //
 // This file is part of SeqArray.
 //
@@ -288,4 +288,36 @@ COREARRAY_DLL_EXPORT SEXP FC_GDS2SNP(SEXP geno)
 	return geno;
 }
 
+
+// ======================================================================
+// SeqArray GDS --> Dosage GDS
+// ======================================================================
+
+static int FC_Num_Sample = 0;
+
+COREARRAY_DLL_EXPORT SEXP FC_SetNumSamp(SEXP num)
+{
+	FC_Num_Sample = Rf_asInteger(num);
+	return R_NilValue;
+}
+
+COREARRAY_DLL_EXPORT SEXP FC_GDS2Dosage(SEXP dosage)
+{
+	int n = LENGTH(dosage);
+	if (n < FC_Num_Sample)
+	{
+		dosage = NEW_NUMERIC(FC_Num_Sample);
+		double *dst = REAL(dosage);
+		for (int i=0; i < FC_Num_Sample; i++)
+			dst[i] = R_NaN;
+	} else if (n > FC_Num_Sample)
+	{
+		double *src = REAL(dosage);
+		dosage = NEW_NUMERIC(FC_Num_Sample);
+		double *dst = REAL(dosage);
+		memcpy(dst, src, sizeof(double)*FC_Num_Sample);
+	}
+	return dosage;
+}
+
 } // extern "C"